U.S. patent application number 16/103441 was filed with the patent office on 2019-01-03 for enhancement using analytics based on vehicle kinematic data.
The applicant listed for this patent is SPEEDGAUGE, INC.. Invention is credited to Christopher J. Carver, Kevin Wolsey.
Application Number | 20190005588 16/103441 |
Document ID | / |
Family ID | 54265472 |
Filed Date | 2019-01-03 |
![](/patent/app/20190005588/US20190005588A1-20190103-D00000.png)
![](/patent/app/20190005588/US20190005588A1-20190103-D00001.png)
![](/patent/app/20190005588/US20190005588A1-20190103-D00002.png)
![](/patent/app/20190005588/US20190005588A1-20190103-D00003.png)
![](/patent/app/20190005588/US20190005588A1-20190103-D00004.png)
![](/patent/app/20190005588/US20190005588A1-20190103-D00005.png)
![](/patent/app/20190005588/US20190005588A1-20190103-D00006.png)
![](/patent/app/20190005588/US20190005588A1-20190103-D00007.png)
![](/patent/app/20190005588/US20190005588A1-20190103-D00008.png)
United States Patent
Application |
20190005588 |
Kind Code |
A1 |
Carver; Christopher J. ; et
al. |
January 3, 2019 |
ENHANCEMENT USING ANALYTICS BASED ON VEHICLE KINEMATIC DATA
Abstract
A computer-implemented method includes receiving from a
telemetric apparatus carried by a first vehicle an identity of an
operator of the first vehicle and kinematic data characterizing
movement and first location of the first vehicle. At least one
weather condition associated with the first location is received
from a source of environmental data. At least one of terrain
geometry of the first location, road speed limit at the first
location, vehicle-to-infrastructure (V2I) data generated by an
instrument proximate the first location, and vehicle-to-vehicle
(V2V) data generated by an instrument proximate the first location
is received from a source of location-specific data. The data
received by the first communication device is stored in a database.
Data from the database is provided to at least one entity.
Inventors: |
Carver; Christopher J.;
(Bainbridge Island, WA) ; Wolsey; Kevin; (Calgary,
CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
SPEEDGAUGE, INC. |
SAN FRANCISCO |
CA |
US |
|
|
Family ID: |
54265472 |
Appl. No.: |
16/103441 |
Filed: |
August 14, 2018 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
14687099 |
Apr 15, 2015 |
10049408 |
|
|
16103441 |
|
|
|
|
61979972 |
Apr 15, 2014 |
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06Q 40/08 20130101;
G07C 5/008 20130101 |
International
Class: |
G06Q 40/08 20060101
G06Q040/08 |
Claims
1. A method for enhancing positioning accuracy in a vehicle, the
method comprising: storing an identity of an operator of a vehicle;
receiving kinematic data characterizing movement of the vehicle
from a telematics device at the vehicle; receiving a location of
the vehicle from the telematics device; generating a driver safety
index corresponding to the identity of the operator of the vehicle
based on the kinematic data characterizing the movement of the
vehicle; generating one or more offsets based on the driver safety
index; and enhancing an accuracy of the location of the vehicle by
applying the one or more offsets to modify the received location of
the vehicle.
2. The method of claim 1, wherein the location of the vehicle is
determined using a Global Navigation Satellite System (GNSS)
receiver.
3. The method of claim 1, wherein the location of the vehicle is
determined using location specific network information
corresponding to one or more wireless telecommunications
networks.
4. The method of claim 1, wherein the location of the vehicle is
determined using location specific network information
corresponding to one or more wireless local area networks.
5. The method of claim 1, wherein the kinematic data characterizing
the movement of the vehicle corresponds to at least one trip
occurring from an engine of the vehicle turning on to the engine of
the vehicle turning off.
6. The method of claim 1, wherein generating the driver safety
index based on the kinematic data characterizing the movement of
the vehicle includes generating the driver safety index based on
the one or more times at which the vehicle was speeding.
7. The method of claim 1, wherein generating the driver safety
index based on the kinematic data characterizing the movement of
the vehicle includes generating the driver safety index based on
the one or more times at which the lateral acceleration of the
vehicle exceeded the predetermined threshold level for more than
the predetermined threshold time.
8. The method of claim 1, further comprising: receiving, kinematic
data characterizing movement of the second vehicle from a second
telemetric device carried by a second vehicle,; and modifying the
kinematic data characterizing the movement of the vehicle based on
the kinematic data characterizing the movement of the second
vehicle.
9. The method of claim 1, further comprising: generating a score
based at least on the kinematic data characterizing the movement of
the vehicle; identifying a deviation between the score and a norm
that corresponds to a plurality of similar vehicles; and
dynamically adjusting an insurance pricing based on the deviation,
wherein the insurance pricing corresponds to the vehicle.
10. The method of claim 9, further comprising identifying that the
deviation between the score and the norm exceeds a threshold,
wherein the threshold is based on a standard deviation
corresponding to the norm.
11. The method of claim 1, further comprising authenticating the
identity of the operator based on biometric data from a biometric
sensor of the telemetric device.
12. A system for enhancing the accuracy of positioning, the system
comprising: a communication interface that receives a location of
the vehicle and kinematic data characterizing movement of the
vehicle from a telemetric device at the vehicle; a memory storing
instructions and an identity of an operator of the vehicle; and a
processor, wherein execution of the instructions by the processor:
generates a driver safety index corresponding to the identity of
the operator of the vehicle based on the kinematic data
characterizing the movement of the vehicle, generates one or more
offsets based on the driver safety index, and enhances an accuracy
of the location of the vehicle by applying the one or more offsets
to modify the location of the vehicle.
13. The system of claim 12, further comprising a temporal database
storing the kinematic data characterizing movement of the vehicle
as a plurality of events, wherein generating the driver safety
index based on the kinematic data characterizing the movement of
the vehicle uses the temporal database to bias toward more recent
events of the plurality of events.
14. A method for vehicle analytics and feedback, the method
comprising: storing information identifying a plurality of road
segments; receiving vehicle traffic data corresponding to the
plurality of road segments and movement of a plurality of vehicles;
receiving a plurality of locations of the first vehicle along a
first road segment of the plurality of road segments from a
telemetric device carried by a first vehicle; identifying one or
more exceedances by the first vehicle based on the plurality of
locations, each of the one or more exceedances indicating a time at
which an aspect of vehicle driving exceeded one or more thresholds;
ranking the first vehicle against the plurality of vehicles based
on exceedances by road segment; generating a driver safety index
for an operator of the first vehicle based at least on ranking the
first vehicle against the plurality of vehicles; and providing risk
feedback based on the driver safety index.
15. The method of claim 14, wherein one or more exceedances
represent one or more times at which the first vehicle exceeded at
least one of one or more speed limits.
16. The method of claim 14, wherein the one or more exceedances
represent one or more times at which a lateral acceleration of the
first vehicle exceeds a lateral acceleration threshold for more
than a threshold amount of time.
17. The method of claim 14, wherein the one or more exceedances
represent one or more times at which a change in heading of the
first vehicle exceeded a heading change threshold.
18. The method of claim 14, wherein the one or more exceedances are
each based on an extent to which a sensor data from a sensor
exceeds a sensor data threshold.
19. The method of claim 14, wherein the risk feedback corresponds
to an adjustment to an insurance pricing corresponding to the first
vehicle.
20. The method of claim 14, wherein ranking the first vehicle
against the plurality of vehicles based on exceedances by road
segment includes identifying that the first vehicle deviates from a
norm.
Description
PRIORITY CLAIM AND CROSS REFERENCE TO RELATED APPLICATION
[0001] This patent application is a continuation and claims the
priority benefit of U.S. patent application Ser. No. 14/687,099,
filed Apr. 15, 2015 and issuing as U.S. Pat. No. 10,049,408, which
claims the priority benefit of U.S. provisional application number
U.S. Prov. Appl. No. 61/979,972 filed Apr. 15, 2014, the
disclosures of which are incorporated herein by reference.
BACKGROUND OF THE INVENTION
[0002] Of late, a category of insurance referred to as usage-based
insurance ("UBI") has been gaining popularity as a method for
insurance companies to more fairly allocate risk among their
customers. The principals of UBI suggest an individual will share
all vehicle usage or exposure data without regard to privacy, such
as location, miles driven by time of day, time of day, and
speeding/braking data all collected from their vehicle using an
onboard device, or a combination of an onboard device and portable
transmitter (i.e., telemetry device) directly or indirectly
controlled by the insurance carrier. The methods of retrieving data
from these sensors for use in the underwriting and modeling of
rates do not discriminate among those driving the vehicle by either
percentage of time or variation in driving scores. Furthermore,
little effort is made to provide contextually relevant data and
information back to the user to improve driving behavior in
real-time.
[0003] The marriage of onboard vehicle performance data with off
board environmental data, replete with GPS satellite-based
determinations of vehicles dynamics, has been shown to have a
significant impact on determination of the frequency of collisions
and near-collisions. To date these contextual computations have
been performed off-board and delivered to the insurance carriers as
a proxy for collisions. Insurance carriers are growing increasingly
concerned about creating these analytics off-board because of three
trends (1) location privacy protection for the insured, (2) an
inability to directly measure and index improvements safety systems
on the vehicles are creating, resulting in more and more
near-collisions as demonstrated by the continued decline in highway
fatalities, and (3) the quantity of data demands more distributed
computing. These factors have created a paradigm shift in risk
management to improve both the individual driver's ability to
receive, and the insurance carrier's ability to offer, loss
prevention services. The growing consensus among safety managers
has been that analyzing the data from historical driving
behavior--or even current driving behavior--has limited value
unless one has a simultaneous (i.e., synchronous) data set from
other drivers sharing the road in a common driving environment.
Developing a portable safe-driving score as an input to a
risk-transfer method, or risk-management device, without the
ability to measure the driver and vehicle response to the forward
assessment of risk (such as a warning to slow down because the
vehicle is about to cross a bridge with known icing conditions) has
been demonstrated to not correlate with negative outcomes (i.e.,
actual loss experience).
[0004] Several embodiments of scoring methods have been documented
in the related patents directed to UBI. Each such method results in
a determination of risk based on granular driving history data
accumulated for specific road segments. These scoring methods
assume a consistent or synchronous data feed from each vehicle, and
do not assume asynchronous methods of data collection.
[0005] Conventional UBI approaches risk management as a definitive
profile, from the onboard device, that does not change much with
environment, or psychographics. The traditional approach has been
to depend on demographics. This simplistic approach provides not
only a generalized assessment of the associated exposure of the
vehicle, plus an unfair bias formed from historic differences by
age, sex and race--without regard for dynamic changes in the
environment, or behavior of the driver as a disconnected entity
from the vehicle. This approach errs on the side of legacy
techniques biased by a driver's demographic history rather than
present behavior, or geographic region. Further, conventional
methods of risk management do not provide consumers choice in
privacy settings, temporary or permanent, without reverting to
prohibitively expensive risk-transfer methods. Thus, UBI as
presently practiced fails to take advantage of the latest advances
in onboard safety systems and behaves in a manner similar to
conventional risk transfer when it fails to reveal transient
changes in driving style. Because it can be shown that in a
majority of cases these transients are of a shorter duration than
the policy period, they are often minimally, or never, reflected in
traditional motor insurance premiums with, or without UBI modifiers
applied.
SUMMARY OF THE INVENTION
[0006] In an embodiment, a computer-implemented method includes
receiving from a telemetric apparatus carried by a first vehicle an
identity of an operator of the first vehicle and kinematic data
characterizing movement and first location of the first vehicle. At
least one weather condition associated with the first location is
received from a source of environmental data. At least one of
terrain geometry of the first location, road speed limit at the
first location, vehicle-to-infrastructure (V2I) data generated by
an instrument proximate the first location and vehicle-to-vehicle
(V2V) data generated by a mobile instrument proximate the first
location is received from a source of location-specific data. The
data received by the first communication device is stored in a
database. Data from the database is provided to at least one
entity.
BRIEF DESCRIPTION OF THE DRAWINGS
[0007] The foregoing aspects these novel methods for dynamic risk
management, along with the many attendant advantages of this
technology, will become more readily appreciated, and the same
better understood, by reference to the following detailed
description and the accompanying drawings, wherein:
[0008] FIG. 1 is a data flow diagram of an archetypical system
demonstrating one exemplary embodiment of the FAIR score
computation and verification system 100;
[0009] FIG. 2 describes a sequential approach to capturing and
scoring the data for Driver, Vehicle, Camera and Infrastructure
data outlined in 105 of FIG. 1;
[0010] FIG. 3 describes a simultaneous process for the verification
of data to be shared either as identifiable, or anonymous data, or
not at all (private data based on a "user key" authentication
process, initiated by the process 100 of FIG. 1 to the user for
authorization to segment each data element by classification into
one of three basic categories--identifiable (shared), anonymous
(shared) or private (not shared);
[0011] FIGS. 4 thru 5 are example diagrams illustrating embodiments
of a detailed method for constructing, normalizing and displaying
scoring indices for an individual risk identified by VIN in the
results database 101 of FIG. 1, within a fleet or group of VINs, or
by decile for an entire group of businesses, for distribution to
internal or external risk managers 150 of FIG. 1;
[0012] FIG. 6 provides an example heat map diagraming the
concentration of road activity, with the highest travel activity
occurring in the regions color coded red and the lowest in the
regions color coded blue;
[0013] FIG. 7 is a schematic view of an exemplary operating
environment in which an embodiment of the invention can be
implemented; and
[0014] FIG. 8 is a functional block diagram of an exemplary
operating environment in which an embodiment of the invention can
be implemented.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
[0015] This patent application is intended to describe one or more
embodiments of the present invention. It is to be understood that
the use of absolute terms, such as "must," "will," and the like, as
well as specific quantities, is to be construed as being applicable
to one or more of such embodiments, but not necessarily to all such
embodiments. As such, embodiments of the invention may omit, or
include a modification of, one or more features or functionalities
described in the context of such absolute terms.
[0016] Embodiments include a method and system for communicating
connected vehicle data for risk management from real-time
asynchronous sources, of different reporting intervals. A fleet of
vehicles with dissimilar connected vehicle hardware, provided by
different service providers, can be combined to form a risk
management data network and risk transfer service for private
passenger vehicles, shared transport and commercial vehicles. In
addition, indexing methods allows for and anticipates the exclusion
of blocks of private information for individual drivers. The
methods provide for a means to keep individual driver, shared
transport, or selective fleet vehicle information anonymous from
the risk managers without impacting the integrity of the resultant
safety systems or compromising onboard safety.
[0017] To clarify, these asynchronous data feeds are not
transmitted with the same periodicity, or even governed by a common
clock, as would occur in a synchronous data feed. The asynchronous
method evaluates, peer to peer, data sets that have not been
previously coordinated in any manner, with the exception of a
common geospatial reference, and achieves the same indexed result
as a synchronous method. The asynchronous method works by
oversampling the environment for three orthogonal indices and
constraining the possible data excursions for two with an apriori
knowledge of the vehicles performance limitations. The resultant
intersection of the three orthogonal indices is a singular, unique
scored result which is therefore portable between reference frames
(vehicles within a fleet or business class) or driven environments
(rating territories). Because the driver is common within two of
the solution sets, there are multiple pairs of indices where a
driver can achieve the same overall Safe-Driving score and thus it
is not correct to say an individual's driver safety index (108) is
portable without the other two indices, vehicle risk (VRI) 109 and
Collision Level (CLI) 110.
[0018] The subject matter of an embodiment is generally related to
the development of a safe driving score for risk management and
risk analytics purposes with intent towards transparency and
portability. More particularly, the use of authenticated
asynchronous data sets from the same driver of different vehicles
from which data is accessible produces a quantifiable index that
reduces a family of conceptual risk predictions to a single
quantifiable, transparent and consistent score.
[0019] Although traffic safety generally reflects the consideration
and integration of three components--the roadway, the vehicle, and
the roadway user--the information needs of roadway users are very
often neglected in traditional driving assessment computations by
risk transfer practitioners. To improve the usability of said score
and prevent compromising the privacy of the driver, a set of
antecedents segment risks based on actual driving results without
divulging location. An optional advantage of an embodiment lies in
its ability to balance a priori knowledge of spatiotemporal
constraint violations, with the a posteriori knowledge of sensor
exceedances, near-collisions and collisions. The exceedances are
defined as the extent to which the sensor signal and noise exceeded
a threshold or limit. This quantification of severity in risk
taking, not simply the frequency of the occurrences, enables
apportionment of risk consistent with a driver's abilities, within
a driving environment and for a defined vehicle from asynchronous
vehicle and sensor data. An embodiment enhances safety and
convenience by providing (1) the ability of vehicle safety systems
to optimize the number of spatiotemporal constraint violations
(such as in-vehicle real-time feedback when driving too fast for a
given condition), (2) the ability to assure privacy when the driver
agrees to share data with an entity or agency (both in the vehicle
and on a mobile device), and (3) facilitating intra-vehicle
portability of driver risk profiles to facilitate risk management
and transfer.
[0020] A plurality of data-logging devices is used by motorists and
motor vehicles on today's highways. The data (referred to as "log
files") from these devices can be shown to be very predictive of
risk in aggregate, and at an individual vehicle or driver level,
regardless of whether the vehicle is itself connected, or the data
comes from a mobile terminal communicating with an onboard device.
The concept of a connected vehicle, one which collects and shares
homogenous, spatiotemporal (time and location stamped) data over a
telemetry network in real-time is a well-established art. Despite
the long standing capability to collect such time-series data, the
ability to gather log files, normalize them with environmental
conditions and vehicle performance, and evaluate driver behavior
(e.g., as low-risk, versus high-risk) has been elusive to the risk
management industry. While some methods have been suggested in
other patents, none describe how to (1) maximize the accuracy of
onboard hardware sensor data collection, (2) deliver onboard usable
peer-based analytics and (3) analyze data from different vehicles
having dissimilar sensors. In summary, an innovation of an
embodiment described herein includes the asynchronous data delivery
and collection methods for the same driver in different vehicles to
achieve a normalized result.
[0021] An embodiment described herein provides a clear method to
normalize the aforementioned heterogeneous, asynchronous data
sources. These normalized results, when segmented by class of
business for commercial risk management, or geographically for
personal safety management, can be shown to have considerable value
for both traditionally insured (guaranteed cost) and the
self-insured fleet.
[0022] An embodiment uses a plurality of analytics (from different
vehicles, sensor types, and sources) predicated on the existence of
one or more Global Navigation Satellite System (GNSS) receiver
(using, for example, the GPS, GLONASS, Galileo or Beidou systems)
capable of time stamping spatiotemporal data. In turn this data, in
a two-way exchange with server side generated risk variables, can
provide contextually relevant and peer-based driver feedback
thereby capturing the true ability of the driver to avoid
collisions or risk exceedances.
[0023] An embodiment advances the art of risk management by
utilizing the combined intelligence of a larger off-board data set,
deep-learning methods of event correlation and myriad nationwide
real-time data (including but not limited to wireless transmission)
sources to describe a predictive algorithm for dynamic risk rating
based on variations in both the vehicle performance and driver
score against the predicted norm for the environmental conditions.
Further, a method of creating and rating the risk profile by the
FAIR (forward-assessment of indexed risk) score method allows for
dynamic changes in the quantity of driving data authorized by the
"opt-in" policy assigned to the component data without invalidating
the overall source. For example the insurer can still rate the risk
even if one driver on the policy chooses to only share mileage
related information and no locations, and another elects to share
considerably more data. The absolute values of the driver score so
computed remains only an input variable for insurance rating and
thus a proxy, albeit ten times more accurate than anything used to
date, for collision frequency. True driving behavior over time is
dynamic, and it can be shown that resolving the data deficiency
created by relative changes in behavior or risk tolerance can be
predicted using three indices--Driver (skill) Safety, Vehicle
(performance) Risk and Collision (frequency) Level. Dynamically
mining variables to compute these indices results from a machine
learning algorithm to detect changes in driver behavior, an
important variation in the dynamic allocation of peril analytics.
Localized comparative rating is the recommended embodiment of the
FAIR score method and not the computation of an absolute driver
score without location adjustments as previous methods suggest.
[0024] In summary, an optionally advantageous differentiator of the
methods asserted and shown, to analyze asynchronous, heterogeneous
data, regardless of the methods used for onboard analysis, data
collection or similarity of data sources, are demonstrable
predictive results which correlate very well with loss history over
time. An important distinction of this three-index predictive
analytics method is the ability to dynamically manage real-time
changes in the risk profile of a driver, an advancement on
traditional UBI risk allocation and underwriting, and described
from here forward as a FAIR score.
[0025] Use of the FAIR Score describes and implies a method to
underwrite or desk-adjust claims, based on the actual risk incurred
relative to the real-time environment, and larger comparative,
dynamically changing environment and population in the locale where
the vehicle is being operated. This approach has great value
because of its ability to produce predictive analytics for
underwriting; more than just the risk of physical damage to a
claimed vehicle, the predictive value can aid when adjusting rates
for bodily injury, property and liability based on an intimate
knowledge of the location and other vehicles in the area.
[0026] For a true usage based underwriting advantage, on any line
of business, the system described here provides the ability to not
only calculate but adjust the earned premium due based on a
normalized score for the environment which varies in a non-linear
fashion. Thus, earned premium, as a percentage of written premium,
varies in a non-linear fashion with dynamically predicted risk
acting as the driver for base rating model. An optionally
advantageous aspect of the usage based method of underwriting lies
in the existence of a process capable of gathering exposure
information from a wide network of sensors, including video, both
inside and outside the vehicle. Because of the nature of the FAIR
score algorithms and the method described, it is not inherently
necessary to synchronize the data received from sensors located
outside a vehicle before the individual risk manager creates a
model to quantify the risk using the three indices derived by
predictive analytic methods described below herein.
[0027] Finally, to utilize the results of these methods may presume
and may require the ability by the insurer to combine and analyze a
statistically significant percentage of similar vehicle traffic
(>5% of the all vehicles) from which the deviations from the
norm can be computed. The normalization process produces a premium
modifier, debit or credit, which when weighted with a FAIR score
can be applied in lieu of the traditional static rating territory,
to dynamically adjust the insured vehicle's risk transfer
pricing.
[0028] A method of predicting risk from the method described herein
does not attempt to define a final risk score, but instead provide
a set of indices through predicative analytics that can be used by
risk managers in at least one of two ways--(1) create in-house risk
models for retained risk and (2) used directly by insurance
underwriters to improve pricing precision.
[0029] FIG. 7 illustrates an example of a computing system
environment 700 in which an embodiment of the invention may be
implemented. The computing system environment 700, as illustrated,
is an example of a suitable computing environment; however it
should be appreciated that other environments, systems, and devices
may be used to implement various embodiments of the invention as
described in more detail below.
[0030] Embodiments of the invention may be implemented in hardware,
firmware, software, or a combination of two or more of each.
Embodiments of the invention may be operational with numerous
general-purpose or special purpose computing system environments or
configurations. Examples of well-known computing systems,
environments, and/or configurations that may be suitable for use
with embodiments of the invention include, but are not limited to,
personal computers, server computers, hand-held or laptop devices,
multiprocessor systems, microprocessor-based systems, set-top
boxes, programmable consumer electronics, network PCs,
minicomputers, mainframe computers, distributed computing
environments that include any of the above systems or devices, and
the like.
[0031] Embodiments of the invention may be described in the general
context of computer-executable instructions, such as program
modules being executed by a computer. Generally, program modules
include routines, programs, objects, components, data structures,
etc. that perform particular tasks or implement particular abstract
data types. Embodiments of the invention may also be practiced in
distributed-computing environments where tasks are performed by
remote processing devices that are linked through a communications
network. In a distributed computing environment, program modules
may be located in both local and remote computer storage media
including memory storage devices.
[0032] With reference to FIG. 7, an exemplary system for
implementing an embodiment of the invention includes a processing
device, such as computing device 700. The computing device 700
typically includes at least one processing unit 702 and memory
704.
[0033] Depending on the exact configuration and type of computing
device, memory 704 may be volatile (such as random-access memory
(RAM)), nonvolatile (such as read-only memory (ROM), flash memory,
etc.) or some combination of the two. This most basic configuration
is illustrated in FIG. 7 by dashed line 706.
[0034] Additionally, the device 700 may have additional features,
aspects, and functionality. For example, the device 700 may include
additional storage (removable and/or non-removable) which may take
the form of, but is not limited to, magnetic or optical disks or
tapes. Such additional storage is illustrated in FIG. 7 by
removable storage 708 and non-removable storage 710. Computer
storage media includes volatile and nonvolatile, removable and
non-removable media implemented in any method or technology for
storage of information such as computer-readable instructions, data
structures, program modules or other data. Memory 704, removable
storage 708 and non-removable storage 710 are all examples of
computer storage media. Computer storage media includes, but is not
limited to, RAM, ROM, EEPROM, flash memory or other memory
technology, CD-ROM, digital versatile disks (DVD) or other optical
storage, magnetic cassettes, magnetic tape, magnetic disk storage
or other magnetic storage devices, or any other medium which can be
used to store the desired information and which can be accessed by
device 700. Any such computer storage media may be part of device
700.
[0035] The device 700 may also include a communications connection
712 that allows the device to communicate with other devices. The
communications connection 712 is an example of communication media.
Communication media typically embodies computer-readable
instructions, data structures, program modules or other data in a
modulated data signal such as a carrier wave or other transport
mechanism and includes any information delivery media. The term
"modulated data signal" means a signal that has one or more of its
characteristics set or changed in such a manner as to encode
information in the signal. By way of example, the communication
media includes wired media such as a wired network or direct-wired
connection, and wireless media such as acoustic, radio-frequency
(RF), infrared and other wireless media. The term computer-readable
media as used herein includes both storage media and communication
media.
[0036] The device 700 may also have an input device 714 such as
keyboard, mouse, pen, voice-input device, touch-input device, etc.
Further, an output device 716 such as a display, speakers, printer,
etc. may also be included. Additional input devices 714 and output
devices 716 may be included depending on a desired functionality of
the device 700.
[0037] According to one or more embodiments, the combination of
software or computer-executable instructions with a
computer-readable medium results in the creation of a machine or
apparatus. Similarly, the execution of software or
computer-executable instructions by a processing device results in
the creation of a machine or apparatus, which may be
distinguishable from the processing device, itself, according to an
embodiment.
[0038] Correspondingly, it is to be understood that a
computer-readable medium is transformed by storing software or
computer-executable instructions thereon Likewise, a processing
device is transformed in the course of executing software or
computer-executable instructions. Additionally, it is to be
understood that a first set of data input to a processing device
during, or otherwise in association with, the execution of software
or computer-executable instructions by the processing device is
transformed into a second set of data as a consequence of such
execution. This second data set may subsequently be stored,
displayed, or otherwise communicated. Such transformation, alluded
to in each of the above examples, may be a consequence of, or
otherwise involve, the physical alteration of portions of a
computer-readable medium. Such transformation, alluded to in each
of the above examples, may also be a consequence of, or otherwise
involve, the physical alteration of, for example, the states of
registers and/or counters associated with a processing device
during execution of software or computer-executable instructions by
the processing device.
[0039] As used herein, a process that is performed "automatically"
may mean that the process is performed as a result of
machine-executed instructions and does not, other than the
establishment of user preferences, require manual effort.
[0040] Referring now to FIG. 8, an embodiment of the present
invention may take the form, and/or may be implemented using one or
more elements, of an exemplary computer network system 800. The
system 800 includes an electronic client device 810, such as a
personal computer or workstation, tablet, smart phone or other
device that may include a processing element, that is linked via a
communication medium, such as a network 820 (e.g., the Internet),
to an electronic device or system, such as a server 830. The server
830 may further be coupled, or otherwise have access, to a database
840 and a computer system 860. Although the embodiment illustrated
in FIG. 8 includes one server 830 coupled to one client device 810
via the network 820, it should be recognized that embodiments of
the invention may be implemented using one or more such client
devices coupled to one or more such servers over one or more such
networks.
[0041] The client device 810 and the server 830 may include all or
fewer than all of the features associated with the device 700
illustrated in and discussed with reference to FIG. 7. The client
device 810 includes or is otherwise coupled to a computer screen or
display 850. The client device 810 may be used for various purposes
such as network- and local-computing processes.
[0042] The client device 810 is linked via the network 820 to
server 830 so that computer programs, such as, for example, a
mobile or other application, running on the client device 810 can
cooperate in two-way communication with server 830. The server 830
may be coupled to database 840 to retrieve information therefrom
and to store information thereto. Database 840 may have stored
therein data (not shown) that can be used by the server 830 to
enable performance of various aspects of embodiments of the
invention. Additionally, the server 830 may be coupled to the
computer system 860 in a manner allowing the server to delegate
certain processing functions to the computer system. In an
embodiment, the client device 810 may bypass network 820 and
communicate directly with computer system 860.
[0043] FIG. 1 illustrates a system 100 according to an embodiment
of the invention, and the elements illustrated in FIG. 1 may be
identical, or otherwise function in a manner similar, to elements
described above with reference to FIGS. 7 and 8. Still referring to
FIG. 1, the system 100 includes onboard devices configured to
provide telemetric data such as telematics service providers
("TSPs") 103, that collect driving data 105 from individual
vehicles or fleets of vehicles. Alternatively, TSPs 103 may be
third-party providers of telemetric data, which is generated by
units supplied by such providers. Driving data 105 may include
driver identity, position, velocity, yaw/pitch/roll, lateral,
vertical and longitudinal acceleration/deceleration, heading and/or
time. Additionally driving data can include driver inputs such as
throttle position, rate of change of steering commands and use of
automation such as ADAS (advanced driver assistance systems). The
driving data is collected by TSP hardware that may be onboard
vehicles (not shown). The term "vehicle" for the purposes of the
discussion herein means any propelled vehicle regardless of the
medium on which it travels and/or its driver controls.
Additionally, a "vehicle" may include an object carried by another
vehicle. The onboard hardware may include cameras, sensors, factory
installed telematics equipment, an after-market telematics device,
a mobile device (such as a phone or tablet), and the like. Using
these various sensors, including their multiple GNSS sensors,
weighted based on differential readings from peered devices,
creates a filter for vehicle dynamics which can be used to cross
correlate asynchronous location reports. The resultant driving data
not only helps identify a driver associated with driving data but
also provides a smoothing function for the normalization of the
data. In such embodiments, the hardware may include an aftermarket
biometric sensor, a mobile device (such as a phone or tablet), and
the like.
[0044] The system 100 gathers data for use by a Driver Scoring
Processor/Database 107 in calculating a Driver Score to be included
in the Driver Scoring Processor/Database. The Driver Scoring
Processor/Database 107, which may be referred to herein simply as
database 107, may be functionally equivalent or at least similar to
server 830 or computer system 860 discussed above with reference to
FIG. 8. The driving data 105 is forwarded to the Driver scoring
database 107 along with environmental data 104 and location
specific data 106 via communication interfaces/devices, such as a
modem or other conventional network-connection device, as generally
indicated in FIG. 1 by arrows. The location specific data 106
and/or the environmental data 104 may be obtained from government
data sources such as local Department of Transportation ("DOT"),
National Oceanic and Atmospheric Administration ("NOAA") and
National Highway Traffic Safety Administration ("NHTSA") data. The
environmental data 104 may include data describing one or more
conventional weather parameters associated with the location of the
vehicle, such as, for example, temperature, relative humidity,
atmospheric pressure, precipitation type, precipitation level, etc.
The location specific data 106 (from one or more GNSS sensor, or
other radio telemetry system) may include geometry of terrain,
speed limit of the road, vehicle to infrastructure ("V2I")
information, and vehicle to vehicle ("V2V") data exchanges. The
Driver scoring database 107 uses a backwards correlated learning
method of mining temporal data (described below as temporal data
mining using deep learning techniques) to generate a driver safety
index 108, a vehicle risk index and Mileage 109, and a collision
level index 110. A fundamentally unique aspect of the method is the
creation of indices using these data mining techniques for the
detection of events correlated with the three indices [driver
safety (DSI) 108, vehicle risk (VRI) 109 and Collision Level (CLI)
110] (see detailed description below of indices determination).
[0045] The deep learning method, or mining method, described herein
begins with a fundamentally different method of data search. Rather
than a forward convolution scheme, the method searches backwards
from the point of an exceedance for similar data patterns to
correlate and then normalizes these patterns for vehicles of, for
example, equivalent weight, driving environment and regional
profile to derive weights to apply to each of the three indices
(DSI, VRI and CLI) at the vehicle level. On roads where there is
insufficient data on collisions, speeding and incidents to provide
a common reference for the CLI, a convolution scheme based on
historically similar peer vehicles (same weight class and
territory). These indices are then summed for a common group of
trips, or period of time using a time weighted average
(proportional to exposure) to produce the fleet level score stored,
along with the individual scores, by VIN number in the results
database 101, again using a temporal database structure for storing
and summing these elements, to bias the results to the most recent
events, or time series data because it has been shown that recent
events are much more predictive of risk factors than historical
patterns when the pattern sequence is changing. For example, there
may exist a braking pattern for an individual driver that appears
to vary from the norm for a certain vehicle type by a trivial
amount until detection of that pattern with greater frequency in a
new geographic region such as on a sharp incline within a city. The
ability to train the index to add this pattern and reverse mine the
data set and see the time series associated with the pattern adds
an optionally advantageous pattern for a particular driver to allow
detection of the overall change in behavior that ultimately is used
to quantify risk using the trip to trip variations shown in FIG. 4.
Stated slightly differently, driving habits in the same class of
vehicle on a rural road, out of traffic, may be low risk and the
same habit in city traffic with steep terrain could be very high
risk.
[0046] Due to the complexity of identifying and coding events
recorded on audio or video records the reverse pattern recognition
method, described above, provides a very effective method of
searching for specific exceedance events, and then by going back in
the data in reverse chronological order similar events can be
identified for inclusion in the index. Without some a priori
knowledge, or learning, video systems historically miss over 90% of
the driving patterns unless there is a catastrophic result, such as
an actual collision. Overall the results obtained by the FAIR score
method have been shown to improve the effectiveness of driver
training by cutting in half the number of exceedances and the
resulting frequency of corrective actions required in over 100 test
fleets (a sample population of 1,320 drivers).
[0047] Once the risk indices (DSI, VRI and CLI) are computed they
are exposure weighted based on time (time in territory) via
geographic information system (GIS) data computations 112 and
stored both at the vehicle level and at the fleet level in the
results database 101. The nomenclature derives from the
identification methods used by the risk managers 150 to query the
database. Underwriting processes may begin with the identification
of the vehicle and the driver to be insured. Therefore these are
the values used to identify the fleet participants for the policy
rating.
[0048] One or more embodiments, and there are multiple methods to
aggregate the driving data 105 shown in FIG. 1, may be specifically
architected to accommodate the use of many different TSPs, their
respective hardware components (sensors), and reporting rates.
Prior art UBI models depend on identical, synchronized devices
and/or sensors with nearly similar reporting rates. In contrast,
the embodiments described herein allow for the use of data from
different hardware platforms (as found in different makes and
models of vehicles), different vehicle types (including on and
off-road vehicles), different vehicle performance capabilities,
different vehicle operating modes (autonomous, semi-autonomous and
manual) and different methods of making the data personally
identifiable. For example, the driving data 105 may be collected
(i) by factory installed telematics equipment and identify the
driver by manual input by the driver of his/her identity entirely
using the installed electronics, (ii) by an after-market telematics
device using an aftermarket biometric sensor configured to identify
the driver, (iii) by a mobile device (such as a phone or tablet) in
real-time or after the fact with the driver identity derived from
his/her association with the mobile device or (iv) a hybrid
approach that combines any, or all, of the above data collection
and driver identification methods.
[0049] FIG. 2 illustrates a sequential approach according to an
embodiment to capturing and scoring driving data 105 (data for
Driver, Vehicle and Camera). In FIG. 2, the optimal vehicle
telematics data (from telematics service providers ("TSPs")) for
off-board scoring is received at predetermined time intervals,
preferably at intervals fewer than 30 seconds in length. If the
data is scored on-board, the score can be summarized and provided
for an entire trip (Engine On to Engine Off). In an embodiment,
Off-board scoring may require transmission of PVT (position,
velocity, heading and time) data upon detection of predetermined
threshold levels of change in such data such as, for example, when
heading changes greater than 30 degrees, speed changes by more than
4 mph (up or down) or lateral acceleration exceeds 400 mg for more
than 100 mS.
[0050] For the application of FAIR scores to regulated fleets with
commercial drivers monitored for Driver Fitness (e.g. applications
for EKG and glucose levels) higher quality data is required for the
Driver scoring database 107 and the driver safety index 108. Unique
to an embodiment of the FAIR score method is the aggregation and
mining of all data sets from various networks prior to the
application of the location specific data 106 and the environmental
data 104, within the database. Of particular importance to
regulated fleets is the segmentation of data in the scoring
database 107 with attention to privacy settings (illustrated in
FIG. 3).
[0051] From each of the different TSP data sources, the system,
regardless of the embodiment, merges the driving data 105 into the
single scoring database 107 to be normalized with the externally
gathered environmental data 104 and the location specific data 106
from one or more GNSS sensors external to the vehicle referenced
with driving data 105. Among the unique attribute of this merging
process is the fact that it's done in reverse chronological order.
From this data the system 100 produces the three indices: (1) the
driver safety index (DSI) 108, (2) the vehicle risk index and
mileage (VRI) 109, and (3) the collision level index (CLI) 110.
Each index 108-110 has two inputs (i) predictive components of the
three data streams (namely, the driving data 105, the location
specific data 106, and the environmental data 104) and (ii) the
predictability of collision, incident, or other claim worthy event
(such as a broken windshield or hail damage) and one output rating
variable that is exposure weighted and fed to both the results
database 101 and the GIS database 112 as the final FAIR score, and
used by the risk manager 150.
[0052] Driver and Vehicle data is brought together over, for
example, a wireless network 102. In addition to identity
verification, an account key passed from the verification process
in the results database 101 ensures that only agreed-upon data
elements are passed (trip pre-processor) to driving data 105.
Following this pre-processing step, the scaled data is adjusted for
Environmental Data (such as weather conditions) 104, and Location
Specific Data (such as local free flow traffic speed or posted
speed information) 106. From within the trip scoring database 107,
smoothing functions (described below herein) are used to deduce
variations between vehicles, temporal data on the condition of the
vehicle (weight, load status, tire pressure, visibility, etc.);
condition of the road (moisture, surface temperature, dry or
otherwise); road geometry (curvature and inclination);
spatiotemporal traffic density data; weather and lighting
conditions; known hazard areas; potentially fraudulent driving data
and so on. When sensor data on board is available, either biometric
or camera, the vehicle independent variables such as condition of
the driver (e.g., drowsy or alert), and focus of the driver's
attention (e.g. forward or distracted) the driving data (105) can
also be adjusted before computing the a trip level score within the
scoring database 107. The trend of these adjustments, by vehicle,
provides a trip to trip variation measurement. By monitoring the
trip to trip variation of these scores with a process also running
in the scoring database 107, high risk drivers and vehicles are
able to be independently identified.
[0053] The system 100 may offer a risk manager a series of options
in the underwriting of a vehicle. The VIN and Driver verification,
in the results database 101, goes beyond collection of driving
scores and raw data to the implementation of several levels of data
that can be used to simply validate existing rating variables for
an automated underwriting system, such as mileage or time of day;
to the modification of the self-reported mileage and variables for
equipment or safety credits in an underwriting system based on
scheduled debits and credits; or as is often found in larger
fleets, use of the three indices (driver safety, vehicle risk and
Collision) in an existing filed rate, to a full loss rated
underwriting regime. In all three scenarios the existence of the
VIN and Driver verification process, in the results database 101,
provides a real-time update on the operation of the telematics
system and the actual mileage for the risk manager at any time
during the policy period for the assessment of earned premium and
changes in risk profile.
[0054] Personally identifiable information (PII) and location data
received from the TSPs 103 may be only passed to the risk manager
150 by the database API if such data has passed authentication key
checks performed locally within the verification database 101. By
doing the process checks as part of the service, the system 100 can
process thousands of vehicles simultaneously. In an embodiment,
performing the identify verification after the data has been
changed to the correct units and entered into the scoring database
107 may reduce the throughput of the system considerably.
[0055] A further purpose of the VIN and Driver verification in the
results database 101 is to provide to third party risk managers
only that data approved by the vehicle owner (usually also the
business owner, or the individual insured).
[0056] For the simplest fleet underwriting operation, the results
database 101 and a "heat map" illustrated in FIG. 6 showing actual
mileage and radius of operation for the vehicles, computed from the
rms (root mean square) value of trips, with seasonal and daily
variations provides an ability to improve accuracy in rating by
150% or more. Such a map-based algorithm may, in an embodiment,
require that the GIS database 112 be included for exposure
computation purposes in the system 100. For such a simple approach,
it has been shown that the results database 101 is required to
correctly weight each trip with a safety index. Without
verification, such as a biometric authentication, or similar
method, the accuracy of the underwriting decreases. For claims
analytics, it is also important to know the identity of the driver
for bodily injury claims. Experience has clearly shown significant
differences in the soft tissue damage caused by a low or high
impact collision simply based on body type and physical
stature.
[0057] A well understood principal of underwriting states that
identifying vehicular-specific perils requires a fundamental
understanding of exposure. The most commonly accepted method of
establishing exposure is to have an insured vehicle provide mileage
traveled or daily radius of operation. When combined with time of
day and geography of operation (including road type) it is commonly
understood by those skilled in the art to produce risk assessment
variables of value to the risk manager 150. Risk managers with
access to the spatiotemporal data and the referenced indices can
compute a time weighted average risk metric for the more accurate
insurance rate for the described exposure and thus price the policy
more accurately for an individual risk in motion. The simplest
embodiment of this method, without computing a FAIR score for the
fleet, includes a simple structured database of mileage and time of
day that can be queried by knowing the VIN from the results
database 101. Thus just using mileage data and time of day as well
understood basic variables in existing rating capability, an
embodiment allows any conventional insurer to begin practicing the
art of Connected Vehicle Insurance with very little investment or
change to their current filed rates.
[0058] As the risk managers become more facile with a particular
implementation of the method for their segment of the population,
or local region, they will determine which factors to weight or
de-weight in their own risk models. These factors from the scoring
database 107 are divided into their component parts for a
comparative, not absolute, score which the system 100 normalizes as
an index. As mentioned above, these indices include (i) the driver
safety index 108, (ii) the vehicle risk index and Mileage 109
(indexed based on a comparison of all mileage reports for similar
classes of business, or geographic region), and (iii) the collision
index 110. The collision index 110 is applied to adjust for dynamic
changes in driving behavior that can be very predictive of
collisions for certain age groups and individuals with known
medical conditions.
[0059] Within the scoring database 107 some of the factors can be
determined with the driving data 105 from within the subject
vehicle or surrounding vehicles, while others may be determined
from the environment data 104 (e.g., obtained from sensors such as
weather sensors on vehicles sharing the same environment) that can
be used to alert a driver to changes in a road condition that will
dramatically change the values of the trip vehicle risk index 109
and the collision index 110.
[0060] FIG. 1 illustrates an embodiment of the data flow through
the system 100 for one or more vehicles, in conjunction with a
commercial auto or fleet underwriting method. As the population of
connected vehicles increases, among individual vehicle owners, the
same method can be applied; it can use an embodiment that clusters
the insured in fleets by risk factors most likely determined from
mobile telephones, public networks, and social media data. This
embodiment and data available from distracted driving work
envisions the clustering of these virtual fleets by the collision
index 110. (See description of data clustering methods below).
[0061] Over a large group of connected vehicles, usually understood
to mean at least 5,000 or more, the method can demonstrate the
ability to produce Driver, Trip and Collision indexes without
having a perfect data set from each vehicle. In FIG. 4 the method
demonstrates that, for private passenger vehicles, the actual trip
to trip score 107 variation over 12% is expected for 50% or more of
drivers.
[0062] The result of the FAIR score method, over previously
disclosed methods, is the segmentation of the vehicle risk index
109, entirely separated from the driver safety index 108, which
should correct to less than a 5% standard deviation for the top
four or more decile for most segmentations of risk. Commercial
Vehicle Insurers when comparing these results with their own
claims-frequency results will be able to validate the following
example observations. As the standard deviation of the driver's
trip to trip score increases above 20% the Collision Level index
increases by at least 100%. For heavier vehicles (>16,000 lbs
GVW) the standard deviation among the top 50% of drivers is
considerably smaller and the Collision Level index also increases
considerably less. A standard deviation in trip scores of more than
30% correlates very strongly with a collision index above the 50th
percentile and therefore predicts collisions at 5.times. the normal
rate. The claims rates are not universally higher because many of
these collisions are a series of low severity, low speed impacts.
The system 100 posits that given the underlying scores and
normalized index, Connected Vehicle Insurers can predict risk for a
vehicle in motion given a driver's behavior based on mileage,
environment, and location, and independent compute risk for the
vehicle in motion based on equipage, load, or maintenance
condition. In an embodiment, the system 100 is only applicable to a
vehicle in motion. Although the sensory inputs may be the same for
the vehicle risk in the static case for the local environment, the
inputs for injury profile and peril analytics for the driver are
not predictive with the method or environmental factors identified
here.
[0063] Another embodiment of the FAIR score would allow the
commercial vehicle insurers to access the raw data and compute
within their own models each decile for desired geographic regions,
demographics and classes of business; and normalize these results
using aggregate and anonymized values for all the regional vehicles
in a particular business class, or other desired anonymized, custom
groupings of vehicles within a demographic.
[0064] Using any of the described or similar methods, the policy
holder gains the capability to secure individual data elements,
scoring parameters or types of data (e.g., data related to a
particular collision). FIG. 3 demonstrates various steps by which a
user might segment his/her data to restrict its receipt by the
insurance company. The steps illustrated in FIG. 3 also allow the
policy holder to provide different privacy settings for the
different drivers on the same vehicle provided the vehicle is
identified by VIN or serial number.
[0065] The commercial vehicle insurer wishing to use their existing
filings, or program insurance business with less than 1,000
vehicles, can use a sub-set of the method. This sub-set can be
applied by assessing only mileage and radius of operation. FIG. 5
illustrates an embodiment of the sub-method using a manual table or
automated computer connection to collect real time mileage,
determine the percentage of the fleet reporting and show actual
radius of operation. To further improve the accuracy of this
sub-method, FIG. 6 demonstrates a heat map showing the
concentration of operations around a specific area of defined risk
relative to the normal operation of a similar business in the same
geographic area. For the early adopter of the methods a visual
representation provides the risk manager easy access to the data
without the integration of the index directly into the rating
engine of the insurer. Not shown in the graphical version of this
embodiment is the use of V2I and other hazardous road information
to dynamically adjust the peril analytics around the zones of risk.
Unlike traditional static maps drawn by zip code in underwriting
models, the method provided produces a virtual map of each index
108-110 computed for geocoded regions and stored in the GIS
(spatial) database 112 to provide a better assessment of the total
risk at any given time. An additional aspect of the present subject
matter and index approach is the clear delineation of risk
associated with the collision level index 110 for the driving
situation (e.g., rush hour deliveries, congested parking lots and
driving in hail, etc.) from that of a driver (e.g., aggressiveness,
impaired vision, fatigue and other debilitating conditions) in a
driver safety index 108.
[0066] An optionally advantageous benefit of the FAIR score method,
from the insurance carriers' perspective, will be the ability to
determine the peril associated with the growing trend of shared
vehicle usage and ownership. Further, in the case of ride-sharing
liability, with a known distance of travel and vehicle risk index
109, known driver safety index (and identity, when known) 108, a
collision level index 110 will reduce the risk transferred to the
risk manager by utilizing the connected vehicle insurance method.
Another embodiment might be the method of setting rates for car
sharing in Connected Vehicle underwriting; a process better
visualized as `reverse-underwriting` or the process of matching a
risk to a rate, rather than the traditional rate for risk model. In
the case of new drivers for a car sharing operation, the method can
use the large data set in the scoring database 107, and the
computed vehicle risk index 109 and the collision level index 110
to quickly arrive at a driver safety index 108, usually within a
single trip. The FAIR score method solves many complicated issues
for current commercial operations attempting to insure, or
self-insure traditional, semi-autonomous and autonomous operations.
The portability of the FAIR score also makes it useful for clients
that regularly move between personal lines and commercial policies
such as the case with Transportation Network Companies (TNCs). For
the risk managers at these operations, this rapid assessment
process reduces collisions and increases the effectiveness of
training programs. Further this process can be utilized to adjust
exposure for drivers who increase their Collision Index by electing
to drive manually, rather than take advantage of the
semi-autonomous or autonomous driving features of their vehicles.
As a final consideration, some businesses will inherently have a
higher collision index due to higher risk operations within high
traffic zones of the city, carry hazardous cargo or operations
during times of the day (such as the late afternoon) when incidents
are higher. The ability to track both the absolute value and the
variation of the index from trip to trip provides the Connected
Vehicle Insurer the ability to improve their rating models with
clear underwriting guidelines, and driver training criteria.
[0067] In the case of completely autonomous vehicles where 100% of
the exposure for the vehicle risk index 109 and Collision Level
index 110 may be allocated to the vehicle manufacturer the
Connected Vehicle Insurer has a nearly indisputable opportunity,
especially with the potential for video support 114, due to the
method, to subrogate any insurance claim by either the insured, or
another claimant. In addition to underwriting decisions, the
ability of insurer to evaluate the driver safety index 108 and
associated limitations of the vehicle risk index 109 in real-time
can also be used to influence other risk control actions.
[0068] For the purposes of insurance claims management by the
Connected Vehicle Insurers the system 100 (as shown in FIG. 1)
receives high resolution crash data that may be part of data 105
from onboard systems and external systems monitoring the general
environment of the vehicle, such as real-time vehicle weather
stations, roadside weather stations, onboard vehicle cameras and
roadside safety cameras. For confidentiality purposes, no more than
20 minutes of data 105 (or video data subset 114) may be included
in an embodiment of the system and combined with a plurality of
data for the claims file. Due to security procedures of the system
100, the driver may have the right to withhold any individual data
set from the claims file, including but not limited to
environmental data, health data, vehicle maintenance and drive
behavior data.
[0069] Description of Temporal Data Mining Using Deep Learning
Techniques:
[0070] Most telematics platforms today, including mobile phones, do
not adapt to the driver's usage patterns, driving style(s), weather
conditions, traffic or interaction with the vehicle. Rather, the
person(s) adapt to the capabilities of the vehicle. To achieve a
vehicle neutral driver safety index 108 and vehicle risk index 109
requires deep learning based on a priori knowledge of how
normalized driver data sets behave. Further, the collision level
index 110 may be calibrated in real time for the environmental data
using a normalized data set for the local region. The system 100
(specifically the data gathering subsystems 104, 105 and 106)
create the primitives that are summarized in the scoring database
107 prior to the creation of the contemplated indices. There is a
plurality of methods for populating the system 100 with results in
the scoring database 107. A unique embodiment of this scoring
database 107 and the primitives selected includes the ability to
utilize deep learning software on the mobile platform, or on
external application to amalgamate the subsystems 104, 105 and 106
into trip data profile (a spatiotemporal representation of the
trip).
[0071] Whether running as a backend service (most common) or as an
onboard service, such as on the vehicle infotainment head unit or
Smart Phone, both process the data similarly prior to its storage
in the database 107. The machine-learning service may require
access to environmental data, location specific risk and traffic
data indexed by vehicle identifier either from a TSP or from a
generally available public service (such as weather data provided
by V2I services). The system 100 accesses several commonly-used
machine adaptation techniques to amalgamate the subsystem data into
a trip score. For example, the database query can provide access to
interfaces for ranking, clustering, classifying, and prediction
techniques. For example, a query of the primitives can provide
most-frequently occurring speeds in similar traffic and weather
conditions as an input to the machine-learning service.
[0072] The resulting data set can also be utilized to infer data
that is not in the database, or cannot be accessed for a particular
trip segment. For example, the application may not have access to
location data for a mobile platform during a trip. However, the
application could request the machine-learning service to predict
the traffic density based on location. The machine-learning service
need only provide the predictions, not the locations, to the driver
safety index creation algorithm. As such, the machine-learning
service can encapsulate the use of sensitive data.
[0073] The scoring database 107 constantly receives and places
input data from multiple sources in archive 113 and/or video
archive 111. The stored input data can be aggregated to discover
features within the data; such as garaging locations based on
actions and times (e.g., Home or Work). The techniques used by the
in-database machine-learning service can also be implemented to
work on the mobile device on a much smaller scale. For example,
aggregated data can be used to train and/or set initial values for
the machine adaptation techniques used by the scoring database
service.
[0074] Exemplary Uses of the Machine-Learning Service Can
Include:
[0075] Predicting a duration of trip before the trip starts, based
on input location, time, other vehicles on the road, prior
behavior, etc.
[0076] Predicting the percentage of time that will be spent
speeding at the onset of the trip, based on input location, time,
traffic, prior behavior, etc.
[0077] Predicting number of passengers in the vehicle, based on
input location, time, music selection, prior behavior, etc.
[0078] Classifying trip risk profile based on the input maintenance
status of the vehicle at the start, etc.
[0079] By providing access to the scoring database 107, a number of
machine adaptation techniques designed to operate on and learn
about user behavior can be enabled either on the backend 105 or on
the mobile platform 103 to enable contextual machine-learning
services.
[0080] From the scoring database 107, the machine-learning service
can generate an output on at least one feature, or the plurality of
features. An example might include selecting a machine-learning
algorithm to perform the machine-learning operation. In particular
embodiments, selecting the machine-learning algorithm to perform
the machine-learning operation to synthesize a desired input, such
as receiving an environmental input when one is missing but
desired.
[0081] In other embodiments, the received data can include data
related to a video, such as data related to one or more images that
show a posted speed limit that is not available in the standard GIS
Database, or data set 112. The machine learning algorithm performs
the function to mine the incoming data stream for this data in an
on-demand process.
[0082] The machine learning engine for the scoring database 107
performs the machine learning and then classifies, clusters, ranks,
and/or predicts data states for the scoring result from given input
data. Classification of driving variables varies by sampling
frequency of the source data and therefore must be performed using
a plurality of statistical classification techniques, such as, but
not limited to, linear classifiers, support vector machines,
quadratic classifiers, kernel estimation, decision trees, neural
networks, traditional Bayesian networks, hidden Markov models,
binary classifiers, and/or multi-class classifiers.
[0083] Clustering driving data provides an essential feature in the
normalization process of local driving behaviors. Clustering
driving scores involves putting the classified primitives (raw
driving data) with a number N of possible values into one of X
clusters, where X is finite, and where the clusters are not
necessarily pre-determined or continuous in time. Generally, each
data item in a given cluster is more similar to each other than to
data item(s) in other cluster(s), but generally this is because of
the temporal nature of the driving history. For example, a
vehicle's location throughout the day within a geographic region
will cluster at locations recently visited, such as work
location(s), client location(s), and other relevant or expected
location(s). Location clusters can vary from driver to driver, and
are not fleet or business specific; a delivery driver who prepares
product at a home office and then frequents a restaurant may
consider the remote office as a "work location" and the restaurant
as an "entertainment location", but his or her spouse who happens
to work for the restaurant a "work location" and the home office as
a "home location". Although a subtle distinction, the ability to
normalize driving behavior correctly depends in many cases on the
ability to infer the intent of the clustered trips.
[0084] When clustering data prior to ranking it for scoring
purposes, it is often important to look at location specific
network information such as wireless telecommunications networks,
local area networks (WiFi), and other networks to determine if
there are additional identifiers to discriminate or infer the
purpose of the cluster around the origin or destination of the
trip. Further, some context-related data can be received from a
network such as number of other employees at "the office".
[0085] Those skilled in the art will understand that the possible
context signals and/or sources of context information is almost
unlimited, and that other context signals and/or sources of context
information are possible in addition, or in the alternative.
[0086] For the determination of the score by database 107, an
embodiment is chosen consistent with the context of the data
collection process, primarily because identifying context helps
predict commuting patterns and thereby create new predictors of
exposure. For instance, an exemplary Driving Score may store
context signals associated with "leaving for work," "driving to
work," "driving home from work," and "arriving home." By knowing
when these times are inconsistent with similar clustered activity a
deviation in the result can be either tagged, or smoothed.
[0087] Other possibilities for machine learning may also exist with
the very simple end objective of predicting the Driving Score for
the next trip as way of determining significant deviations without
seeding new clusters, and requiring X+1 clusters each time a
significant deviation is detected between the value in the scoring
database 107 and the predicted value in the scoring database.
[0088] Detailed Description of Indices Determination:
[0089] An objective of the index creation process is to create a
predictive model that can accurately identify the occurrence of
events which cause an increase in exposure or risk due to Vehicle
risk (VRI) 109, Driver safety (DSI) 108 and collision level index
(CLI) 110. It is not necessarily the intent of the method to create
a final set of risk analytics from these three indexes; that task
remains the domain of the individual insurance carrier.
[0090] Once the Driving Score for each trip has been computed and
stored 107 the process of mining all the trip data, environmental
data and collision data begins to classify the components of the
three risk related variables, referred to here as the DSI, VRI and
CLI index (108, 109, 110). The CLI is the first to be computed
since it is the dependent variable in the computation of the
VRI.
[0091] The majority of existing temporal classification methods
assume that each input trip score (normally represented by a single
series, but can be represented by multiple time series) is
associated with a uni-variant classification, or profile,
representing behavior for the entire trip. The conventional UBI
models assume the prediction of risk to depend on the historical
sum of events. That is, they assume that all temporal observations
are potentially equally useful for classification and as such are
time-independent. However, the assumption has been shown to not be
suitable when considering event detection problems within the
context of trips where the location and actors (other vehicles,
weather, and pedestrians) are changing.
[0092] Given that trip scores are associated with specific time
points (or time intervals), each time series or trip is affected by
the most recent behavior. Our approach to mining the indices
proposes a novel temporal pattern mining approach for event
detection within the clustered trip scores, this approach takes
into account driver experience levels and hyper-local nature of
driving decisions. Mining temporal patterns backward in time,
starting from trip or behavior patterns as an entire class related
to the most recent trip (and Driver Score 107), results in a trip
classification more valuable than the simple summary of events by
classification within a trip. Incorporating the concept of trip
aging in temporal pattern mining produces results more predictive
of actual loss experience (as measured by collisions) than simply
treating all trips as time independent (both from an aging and a
duration perspective). An objective with the trip level mining, as
compared to the vehicle detail mining, is to determine weights for
indices from the trip duration and scoring patterns that can
accurately predict adverse outcomes for future trips. This result
is extremely useful in loss control or risk management because it
enables reliable prediction of a loss before it occurs, and thereby
provides value-added services such as intelligent driver
monitoring/decision support when faced with adverse driving
conditions.
[0093] The task of creating a predictive model from the indices can
be challenging because they are multivariate and not a coherent
representation of actual exposure. The method described works best
when applied to large samples of data for which peer-to-peer
analytics can be applied and exceedences by road segmented ranked
and stored for analytics during low density traffic periods.
Regardless of traffic density for the trip segment, the same
reverse chronological method of pattern methods described in the
individual trip scores are applied to determine the risk density
functions used to produce the indices found in the results database
101 at the fleet level. The resulting indices are therefore
available to use directly within the fleet's risk management
program or, more likely, as an additional piece of data in a
software application providing constructive driver risk feedback as
an alternative to temporal data in raw forms such as speeding,
cornering and braking.
[0094] An embodiment includes a method of enhancing the inherent
accuracy of the onboard location technology by the application of
driver safety index (DSI) offsets, over the V2I or other radio
telemetry network, for specific driver characteristics to improve
the accuracy of the GNSS signals and improve the response rate of
these inputs to the vehicle's factory equipped or after-market
safety system.
[0095] An embodiment includes a mobile software application,
operating on at least one computing device onboard the vehicle, to
provide contextual driver feedback data (based on the VRI) to
manually or automatically modify the vehicle systems responses
based on peer vehicle sensor exceedances, or prevent the vehicle
from operating in either a semi-autonomous or fully autonomous
mode.
[0096] An embodiment includes receiving, by at least one computing
device, a plurality of asynchronous and heterogeneous digital data
inputs streams from one or more service providers, the data inputs
streams comprising vehicle related data; storing, by the at least
one computing device, the vehicle related data, environmental data,
and location specific data in a scoring database; and, by the at
least one computing device, providing access to the scoring
database to a plurality of insurers to use to set rates for a
connected vehicle insurance. One or more service providers may
comprise at least one of a vehicle OEM, an aftermarket telematics
services provider, and a wireless carrier. The data inputs streams
may include at least one of structured location data, diagnostic
vehicle data, prognostic vehicle data, unstructured driver behavior
data, structured driver behavior data, biometric data, imaged data,
audio data, video data, and signaling data. An embodiment may
further include differentiating, by the at least one computing
device, connected vehicle insurance from usage based insurance by
using asynchronous, mixed resolution data from dissimilar and
sometimes incomplete data sets merged with high resolution
environmental and location specific data to create a risk index for
use by an insurer. The scoring database may comprise a driver
safety data set, and may further comprise segmenting, by computing
device, the driver and vehicle data set into a driver safety index,
vehicle risk index and Collision Level index to be used to predict
the appropriate risk variable for one policy. An embodiment may
include using information from a VIN database and mileage data to
rate an insurance policy using existing filed rating plans. An
embodiment may include using the Collision Index to predict a
likelihood of a collision for fleet vehicles or vehicles assembled
in a virtual fleet based on demographics, and/or local geographic
constraints.
[0097] An embodiment includes a system comprising at least one
computing device configured to perform an insurance claims
management process that provides event data, from within or
external to the vehicle, for insurance claims management by
creating a virtual replay of the event and real-time adjustment
using prescribed data. The insurance claims management process may
use imaging and video data to create an incident or collision
reconstruction from an unsynchronized digital data set.
[0098] An embodiment includes a system comprising at least one
computing device implementing a vehicle trip scoring database
configured to create relative driver safety indices to be used to
predict collisions based on the trip to trip standard deviation of
the results. The relative driver safety indices may be used to
calculate a trip to trip standard deviation based on an individual
driver's fuel economy information as driver behavior data. The
relative driver safety indices may include Vehicle and Collision
indexes, and a driver safety index that is portable between
different Vehicles provided the Vehicle and Collision indexes are
known for an entire duration of exposure. The relative driver
safety indices include a Drivers Index and a Collision index, and
the Collision index comprises monitoring a policy holder or
driver's response to the Drivers Index. The relative driver safety
indices may include a vehicle risk index and exposed mileage and a
Collision index, and the Collision index comprises monitoring the
vehicle risk index and exposed mileage.
[0099] An embodiment includes a system comprising at least one
computing device implementing a database accessible by insurance
carriers, the database storing VINs, organized by geographic
region, wherein the insurance carriers access the database to
identify existing prospects or insured vehicles. The database may
store mileage data, and the VIN and mileage data retrieved can be
utilized to auto populate a rating table. The insurance carriers
may access the database to generate training feedback for a fleet,
based on a retrieved driver safety and Collision index. The
insurance carriers may access the database to generate training
feedback for a shared vehicle driver using a retrieved driver
safety and Collision Index. The insurance carriers access the
database to generate maintenance feedback for a fleet based on a
retrieved vehicle risk and Collision index.
[0100] An embodiment includes a system comprising at least one
computing device storing a plurality of types of data or a
plurality of data sets, the at least one computing device being
configured to allow a policy holder, an individual driver, group of
drivers or fleet to selectively restrict transmission, by a service
provider, of one or more of the plurality of types of data, or one
or more of the plurality of data sets to an risk manager for use in
a driver safety index. At least one computing device is further
configured to uniquely identify a vehicle, or the embedded device,
allows an individual vehicle, shared vehicles, or a fleet of
vehicles to selectively restrict transmission, by the service
provider, of a discriminate one or more of the plurality of data
sets to the risk manager for use in a vehicle risk index. The at
least one computing device or the embedded device is further
configured to uniquely identify an individual vehicle, shared
vehicles, or a fleet of vehicles to selectively restrict
transmission, by the service provider, of a discriminate one or
more of the plurality of data sets to the risk manager for use in a
Collision Index.
[0101] While the preferred embodiment of the invention has been
illustrated and described, as noted above, many changes can be made
without departing from the spirit and scope of the invention.
Accordingly, the scope of the invention is not limited by the
disclosure of the preferred embodiment. Instead, the invention
should be determined entirely by reference to the claims that
follow.
* * * * *