U.S. patent application number 17/306507 was filed with the patent office on 2021-11-11 for system and method for detecting handwriting problems.
The applicant listed for this patent is INVOXIA. Invention is credited to Arthur BELHOMME, Amelie CAUDRON, Fabrice DEVIGE, Eric HUMBERT.
Application Number | 20210345913 17/306507 |
Document ID | / |
Family ID | 1000005786402 |
Filed Date | 2021-11-11 |
United States Patent
Application |
20210345913 |
Kind Code |
A1 |
HUMBERT; Eric ; et
al. |
November 11, 2021 |
System and Method for Detecting Handwriting Problems
Abstract
A method for detecting handwriting problem, comprising:
acquiring, by a handwriting instrument comprising one motion
sensor, motion data while a user is using the handwriting
instrument, analyzing the motion data by an artificial intelligence
trained to detect a handwriting problem.
Inventors: |
HUMBERT; Eric; (BOULOGNE
BILLANCOURT, FR) ; CAUDRON; Amelie; (PARIS, FR)
; BELHOMME; Arthur; (PARIS, FR) ; DEVIGE;
Fabrice; (VANVES, FR) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
INVOXIA |
ISSY LES MOULINEAUX |
|
FR |
|
|
Family ID: |
1000005786402 |
Appl. No.: |
17/306507 |
Filed: |
May 3, 2021 |
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06K 9/224 20130101;
G06K 9/00416 20130101; A61B 5/4088 20130101; A61B 5/7267 20130101;
G06K 9/6256 20130101; G06K 9/00429 20130101; A61B 5/1124
20130101 |
International
Class: |
A61B 5/11 20060101
A61B005/11; G06K 9/00 20060101 G06K009/00; G06K 9/22 20060101
G06K009/22; G06K 9/62 20060101 G06K009/62; A61B 5/00 20060101
A61B005/00 |
Foreign Application Data
Date |
Code |
Application Number |
May 4, 2020 |
EP |
20305430.9 |
Claims
1. A method for detecting handwriting problem, comprising:
acquiring, by means of a handwriting instrument comprising at least
one motion sensor, motion data while a user is using said
handwriting instrument, analyzing said motion data by an artificial
intelligence trained to detect a handwriting problem.
2. The method according to claim 1, wherein the artificial
intelligence is a neural network.
3. The method according to claim 2, the method further comprising a
prior learning step comprising: acquiring a plurality of motion
data from a plurality of persons using said handwriting instrument,
labelizing said acquired data, using end-to-end supervised learning
to train the neural network until it converges, storing said neural
network.
4. The method according to claim 3, wherein the acquired data are
classified in at least one of the following classes: type of grip
on the handwriting instrument, pressure applied on the handwriting
instrument, use of the handwriting instrument among writing,
drawing or coloring, fluidity of writing, dyslexia, dysgraphia,
wrong ductus.
5. The method according to claim 8, further comprising acquiring
vibration data by a stroke sensor, the method further comprising a
prior learning step comprising: acquiring a plurality of motion
data and vibration data from a plurality of persons using said
handwriting instrument, processing the vibration data to obtain
stroke timestamps labels, using supervised learning to train said
neural network until it converges, storing said neural network.
6. The method according to claim 5, wherein the features extracted
from the strokes timestamps comprise: total strokes duration, total
in air stroke duration, strokes mean duration, strokes mean and
peak velocity, number of pauses during use of the handwriting
instrument, ballistic index, which corresponds to an indicator of
handwriting fluency which measures smoothness of the movement
defined by the ratio between the number of zero crossings in the
acceleration and the number of zero crossings in the velocity,
number of zero-crossing in the acceleration during strokes, number
of zero-crossing in the velocity during strokes.
7. The method according to claim 5, wherein the extracted features
of the stroke timestamps are classified in at least one of the
following classes: type of grip on the handwriting instrument,
pressure applied on the handwriting instrument, use of the
handwriting instrument among writing, drawing or coloring, fluidity
of writing, dyslexia, dysgraphia, wrong ductus.
8. The method according to claim 2, wherein the neural network is
further trained with a data base of letters and numbers correctly
formed, a sequence of strokes and a direction of said strokes of
the sequence of strokes being associated to each letter and number
of the data base, and wherein, based on the motion and vibration
data acquired during the use of the handwriting instrument, the
neural network determines if the user is forming letters and
numbers correctly.
Description
[0001] This application claims priority from European patent
application EP20305430.9, filed on 4 May 2020, this content being
incorporated herein by reference.
TECHNICAL FIELD
[0002] This disclosure pertains to the field of systems and methods
for detecting handwriting problems.
BACKGROUND ART
[0003] Currently, several systems or methods exist for detecting
particular handwriting problems. Among these handwriting problems,
one can found dyslexia, learning disabilities, difficulty in
forming letters, etc.
[0004] Currently, the methods used to detect handwriting problems
are based on the use of a stylus-pen with which a user is asked to
write on a mobile terminal specifically dedicated to detecting
handwriting problems. Such method is for example disclosed in the
document U.S. Pat. No. 6,304,667. In this document, the user writes
on a mobile terminal such as an electronic tablet. The tablet
comprises an embedded pressure sensor able to calculate the
pressure of the stylus on the electronic tablet when the user is
writing. Furthermore, the method uses a database of distorted
letters with which the letters written by the user are compared.
This method allows detecting handwriting problems such as dyslexia.
However, the material used to detect the problem is very specific
and does not allow detecting easily handwriting problems for a
large number of people.
SUMMARY
[0005] One purpose of this disclosure is to improve the
situation.
[0006] It is proposed a system for detecting handwriting problems
comprising: [0007] a handwriting instrument including a body
extending longitudinally between a first end and a second end, said
first end having a writing tip which is able to write on a support,
said handwriting instrument further including at least one motion
sensor configured to acquire data on the handwriting of the user
when a user is using said handwriting instrument, [0008] one
calculating unit communicating with said motion sensor and
configured to analyze said data by an artificial intelligence
trained to detect whether said user has handwriting problems.
[0009] This system presents the technical advantage of being able
to detect a handwriting problem with a regular handwriting
instrument. The proposed system is really easy to use since no
further electronic components such as a dedicated tablet and stylus
are needed.
[0010] The following features can be optionally implemented,
separately or in combination one with the others:
[0011] The motion sensor and the calculating unit are embedded in a
second extremity of the handwriting instrument.
[0012] The system further comprises a detection device, said
detection device comprising the motion sensor and the calculation
unit, said detection device being mounted on the second extremity
of the handwriting instrument.
[0013] The system can then be used with any already existing
handwriting instrument.
[0014] The motion sensor is embedded in the handwriting instrument,
the handwriting instrument further included a short-range radio
communication interface configured to communicate raw data acquired
by the motion sensor to a mobile device comprising said calculating
unit via a communication interface of said mobile device.
[0015] The motion sensor comprises two three-axis
accelerometers.
[0016] Therefore, the motion sensor used to detect handwriting
problems does not consume a lot of power.
[0017] The handwriting instrument comprises two motion sensors
being one three-axis accelerometer and one three-axis
gyroscope.
[0018] It increases the precision in the acquired motion data.
[0019] The gyroscope comprises a wake-up input suited for receiving
a wake-up signal from said calculating unit when a movement is
detected by the accelerometer, said gyroscope being configured for
switching into an active state when said wake-up signal is
received.
[0020] This configuration allows to reduce the consumption of power
needed by the system.
[0021] The system further comprises a pressure sensor embedded in
the pen or the pencil, said calculating unit being configured to
further receive data acquired by said pressure sensor.
[0022] The system is then able to detect handwriting problems on
the basis of different kind of data: motion data, pressure data,
etc.
[0023] The system further comprises a stroke sensor configured to
acquire stroke data while the user is using the handwriting
instrument, said artificial intelligence being trained with said
stroke data to determine handwriting problems.
[0024] The artificial intelligence is then able to determine when
the user is actually using the handwriting instrument on a support
and differentiate the data corresponding to an actual use of the
handwriting instrument and the data acquired while the handwriting
instrument is just hold in the air.
[0025] The stroke sensor is the motion sensor.
[0026] Advantageously, the stroke sensor is a pressure sensor, or a
contact sensor, or a vibration sensor.
[0027] Advantageously, the system does not use another embedded
sensor. The system remains compact and low-power consuming.
[0028] In another aspect, it is proposed a method for detecting
handwriting problem, comprising: [0029] acquiring, by means of a
handwriting instrument comprising at least one motion sensor,
motion data while a user is using said handwriting instrument,
[0030] analyzing said motion data by an artificial intelligence
trained to detect a handwriting problem.
[0031] The following features can be optionally implemented,
separately or in combination one with the others:
[0032] The artificial intelligence is a neural network.
[0033] Preferably, the neural network is a deep neural network.
[0034] The method further comprises a prior learning step
comprising: [0035] acquiring a plurality of motion data from a
plurality of persons using said handwriting instrument, [0036]
labelizing said acquired data, [0037] using end-to-end supervised
learning on the labelizing to train the neural network until it
converges, [0038] storing said neural network.
[0039] The neural network can then be trained by means of an
end-to-end classification. This improves the precision of the
results.
[0040] The acquired data are classified in at least one of the
following classes: [0041] type of grip on the handwriting
instrument, [0042] pressure applied on the handwriting instrument,
[0043] use of the handwriting instrument among writing, drawing or
coloring, [0044] fluidity of writing, [0045] dyslexia, [0046]
dysgraphia, [0047] wrong ductus.
[0048] The system can then detect a lot of different and particular
handwriting problems.
[0049] The method further comprises acquiring vibration data by a
stroke sensor, the method further comprising a prior learning step
comprising: [0050] acquiring a plurality of motion data and
vibration data from a plurality of persons using said handwriting
instrument, [0051] processing the vibration data to obtain stroke
timestamps labels, [0052] using supervised learning on the
processing to train said neural network until it converges, [0053]
storing said neural network.
[0054] The neural network can then be trained by segmentation and
classification of strokes. The size of the neural network is then
smaller than the neural network trained according to the end-to-end
classification.
[0055] The features extracted from the strokes timestamps comprise:
[0056] total strokes duration, [0057] total in air stroke duration,
[0058] strokes mean duration, [0059] strokes mean and peak
velocity, [0060] number of pauses during use of the handwriting
instrument, [0061] ballistic index, which corresponds to an
indicator of handwriting fluency which measures smoothness of the
movement defined by the ratio between the number of zero crossings
in the acceleration and the number of zero crossings in the
velocity, [0062] number of zero-crossing in the acceleration during
strokes, [0063] number of zero-crossing in the velocity during
strokes.
[0064] The classification of the features of stroke timestamps is
made by a hand-crafted algorithm or a learned model.
[0065] Two approaches can then be used to extract the features.
[0066] Method according to any of claims 15 to 17, wherein the
extracted features of the stroke timestamps and motion data are
classified in at least one of the following classes: [0067] type of
grip on the handwriting instrument, [0068] pressure applied on the
handwriting instrument, [0069] use of the handwriting instrument
among writing, drawing or coloring, [0070] fluidity of writing,
[0071] dyslexia, [0072] dysgraphia, [0073] wrong ductus.
[0074] The neural network is further trained with a data base of
letters and numbers correctly formed, a sequence of strokes and a
direction of said strokes of the sequence of strokes being
associated to each letter and number of the data base, and wherein,
based on the motion and vibration data acquired during the use of
the handwriting instrument, the neural network determines if the
user is forming letters and numbers correctly.
[0075] The system is then adapted to detect a lot of different
handwriting problems, with a high precision, small number of
components and easy use.
BRIEF DESCRIPTION OF DRAWINGS
[0076] Other features, details and advantages will be shown in the
following detailed description and on the figures, on which:
[0077] FIG. 1 is an illustration of a system for detecting
handwriting problems according to a first embodiment.
[0078] FIG. 2 is a block schema of the system illustrated in FIG.
1.
[0079] FIG. 3 is an illustration of a system for detecting
handwriting problems according to a second embodiment.
[0080] FIG. 4 is a block schema of the system illustrated in FIG.
3.
[0081] FIG. 5 is an illustration of a system for detecting
handwriting problems according to a third embodiment.
[0082] FIG. 5 is an illustration of a system for detecting
handwriting problems according to a fourth embodiment.
[0083] FIG. 7 is an illustration of a system for detecting
handwriting problems according to an alternative embodiment of
FIGS. 1 and 2.
[0084] FIG. 8 is a block diagram illustrated the training phase of
the neural network according to a first embodiment.
[0085] FIG. 8 is a block diagram illustrated the training phase of
the neural network according to a second embodiment,
[0086] FIG. 10A to 10C illustrate block diagrams of the collect
phase, training phase and inference phase of the trained neural
network.
DESCRIPTION OF EMBODIMENTS
[0087] Figures and the following detailed description contain,
essentially, some exact elements. They can be used to enhance
understanding the disclosure and, also, to define the invention if
necessary.
[0088] It is now referred to FIGS. 1 to 7 illustrating embodiments
of a system 1 for detecting handwriting problems. The same
reference numbers are used to describe identical elements of the
system.
[0089] In an embodiment, a handwriting problem which can be
detected according to the present disclosure can be dyslexia,
dysgraphia or a difficulty to reproduce characters.
[0090] FIGS. 1 and 2 generally illustrates a system 1 according to
a first embodiment. The system 1 comprises a handwriting instrument
2. Typically, the handwriting instrument 2 can be a pen, a pencil,
a brush or any element allowing a user to write or draw with it on
a support. Typically, the support can be paper, canvas, or any
surface on which a user can write or draw. The support can also be
a coloring book.
[0091] The handwriting instrument 2 comprises a body 3 extending
longitudinally between a first end 4 and a second end 5. The first
end 4 comprises a writing tip 6 which is able to write on a
support. Typically, the tip 6 can deliver ink or color.
[0092] The handwriting instrument 2 further includes at least one
motion sensor 7, In one embodiment, the motion sensor 7 can be a
three-axis accelerometer or a three-axis gyroscope.
[0093] In the illustrated embodiments on FIGS. 1 to 7, the
handwriting instrument 2 preferably includes two motion sensors 7.
In a preferred embodiment, the handwriting instrument 2 comprises
two three-axis accelerometers. In another preferred embodiment, the
handwriting instrument 2 comprises one three-axis accelerometer and
one three-axis gyroscope.
[0094] The at least one motion sensor 7 is able to acquire data on
the handwriting of the user when the user is using the handwriting
instrument 2. These data are communicated to a calculating unit 8
which is configured to analyze the data and detect an eventual
handwriting problem of the user. The calculating unit 8 can
comprise a volatile memory to store the data acquired by the motion
sensor 7 and a non-volatile memory to store a model enabling the
detection of handwriting problem.
[0095] The handwriting instrument 2 can also comprise a short-range
radio communication interface 9 allowing the communication of data
between the motion sensor 7 and the calculating unit 8. In one
embodiment, the short-range radio communication interface is using
a Wi-Fi, Bluetooth.RTM., LORA.RTM., SigFox.RTM. or NBIoT network.
In another embodiment, it can also communicate using a 2G, 3G, 4G
or 5G network.
[0096] The handwriting instrument 2 further includes a battery 10
providing power to at least the motion sensor 7 when the user is
using the handwriting instrument. The battery 9 can also provide
power to the calculating unit 8 when the calculating unit is
included in the writing instrument 2.
[0097] More specifically, in the embodiment of FIGS. 3 and 4, the
handwriting instrument 2 comprises the at least one motion sensor
7, the short-range radio communication interface 9 and the battery
10. The system 1 further comprises a mobile device 11, distinct
from the handwriting instrument 2, The mobile device 11 can
typically be an electronic tablet, a mobile phone or a computer.
The mobile device 11 comprises the calculating unit 8, The mobile
device 11 further comprises a short-range radio communication
interface 12 enabling communication between the calculating unit 8
and the handwriting instrument 2.
[0098] In this embodiment, the calculating device 8 of the mobile
device receives raw data acquired by the motion sensor 7 and
analyzed them to detect an eventual handwriting problem.
[0099] In another embodiment illustrated FIGS. 5 and 6, the motion
sensors 7, the calculating unit 8, the short-range radio
communication interface 9 and the battery 10, are not embedded in
the handwriting instrument 2. The electronics 20 can be comprised
in a detection device 13, distinct from the handwriting instrument
2. The detection device 13 can be mounted on the second end 5 of
the handwriting instrument 2.
[0100] In this embodiment, the detection device 13 comprises a body
14 to be mounted on the second end 5 of the handwriting instrument
2 and a protuberant tip 15 able to be inserted in the body 3 of the
handwriting instrument 2. Preferably, one motion sensor 7 can be
provided on the protuberant tip 15 and another motion sensor 7 can
be provided in the body 14 of the detection device 13. By this
means, the two motions sensors 7 are able to acquire different data
during the handwriting of the user.
[0101] In another embodiment, the motions sensors 7 are provided in
the body 14 of the detection device 13. By this means, the
detection device 13 can be mounted on any type of handwriting
instrument 2, without necessitating a hollow body 3 of the
handwriting instrument 2.
[0102] In another embodiment illustrated on FIG. 7, the at least
one motion sensor 7, the calculating unit 8, the short-range radio
communication interface 9 and the battery 10 are directly embedded
in the handwriting instrument 2.
[0103] In this embodiment, one motion sensor 7 can be provided
close to the first end 4 of the handwriting instrument 2, while
another motion sensor 7 can be provided on the second end 5 of the
handwriting instrument 2.
[0104] In an embodiment, the handwriting instrument 2 can also
comprise a pressure sensor able to acquire data. These data can be
transmitted to the calculation unit that analyze these data and the
data acquired by the at least one motion sensor 7.
[0105] The pressure sensor can be embedded in the handwriting
instrument 2 or in the detection device 13.
[0106] In all the embodiments described above, the calculating unit
8 receives data acquired from at least on motion sensor 7 and from
the pressure sensor 15, if applicable, to analyze them and detect a
handwriting problem.
[0107] More specifically, the calculating unit 8 can store an
artificial intelligence model able to analyze the data acquired by
the motion sensor 7. The artificial intelligence can comprise a
trained neural network.
[0108] In one embodiment illustrated on FIG. 8, the neural network
is trained according to the method of using intermediate features
extraction.
[0109] More particularly, at step S1, the motion sensor 7 acquires
data during the use of the handwriting instrument 2.
[0110] At step S2, the neural network receives the raw signals of
the data acquired at step S1. The neural network also receives the
sample labels at step S3. These labels correspond to whether or not
the signal corresponds to a stroke.
[0111] More precisely, the neural network is able to determine if
the signal correspond to a stroke on a support. The neural network
is then able to determine stroke timestamps.
[0112] More particularly, this means that the neural network is
able to determine for each stroke timestamps if a stroke has
actually been made on the support by the user during the use of the
handwriting instrument 2.
[0113] At step S4, the calculating unit 8 performs a stroke
features extraction to obtain intermediate features at step S5.
[0114] These intermediate features comprise, but are not limited
to: [0115] total strokes duration, [0116] total in air stroke
duration, [0117] strokes mean duration, [0118] strokes mean and
peak velocity, [0119] number of pauses during use of the
handwriting instrument, [0120] ballistic index, which corresponds
to an indicator of handwriting fluency which measures smoothness of
the movement defined by the ratio between the number of zero
crossings in the acceleration and the number of zero crossings in
the velocity, [0121] number of zero-crossing in the acceleration
during strokes, [0122] number of zero-crossing in the velocity
during strokes.
[0123] From these intermediate features, the neural network is able
to derive indications about handwriting problems.
[0124] At step S6, an algorithm is able to derive indications about
handwriting problems.
[0125] This algorithm can be a learned model such as a second
neural network, or a handcrafted algorithm.
[0126] In the embodiment where a learned model such as a neural
network is used, the model is trained on a supervised
classification task, where the inputs are stroke features with
labels, and the outputs are handwriting problems.
[0127] In the embodiment where a hand-crafted algorithm is used,
the hand-crafted algorithm can compute statistics on the stroke
features and compare them to thresholds found in the scientific
literatures, in order to detect handwriting problems.
[0128] Finally, at step S7, the system is able to detect
handwriting problems. These handwriting problems include but are
not limited to: [0129] dyslexia, [0130] dysgraphia, [0131] wrong
grip of the handwriting instrument, [0132] bad character
writing.
[0133] In another embodiment illustrated on FIG. 9, the neural
network is trained according to the method of end-to-end
classification.
[0134] According to this embodiment, at step S10, the data are
acquired by the motion sensor 7.
[0135] The classification is made in step S11. To do learn the
classification task, the neural network receives the raw signal of
the data acquired by the motion sensor 7 and global labels (step
S12). The global labels corresponds to the handwriting problems to
be detected by the neural network which can be, but are not limited
to: [0136] dyslexia, [0137] dysgraphia, [0138] wrong grip of the
handwriting instrument, [0139] bad character writing.
[0140] In step S13, the neural network delivers the result.
[0141] The trained neural network described in reference with FIGS.
8 and 9 is stored.
[0142] The neural network can be stored in the calculating unit
8.
[0143] FIGS. 10A to 100 illustrate more specifically the embodiment
described with reference to FIG. 8.
[0144] In order of segment the strokes (step S2 of FIG. 8), the
neural network may determine the timestamps of the strokes on the
support.
[0145] This information can be detected by a stroke sensor 16. The
stroke sensor 16 is advantageously embedded in the handwriting
instrument or in the detection device 13 mounted on the handwriting
instrument.
[0146] In an embodiment, the stroke sensor 16 may be a pressure
sensor, a contact sensor or a vibration sensor. Then, the neural
network receives the data collected by the stroke sensor 16 at step
S3.
[0147] In a preferred embodiment illustrated FIGS. 10A to 100, the
stroke sensor 16 is the motion sensor 7. More preferably, the
motion sensor 7 is a three-axis accelerometer.
[0148] FIG. 10A illustrates the collect of data used during the
training phase of the neural network, which is illustrated FIG.
10B, Finally, FIG. 100 illustrates the inference of the neural
network by a user of the handwriting instrument.
[0149] To use the motion sensor 7 as the stroke sensor 16, the
accelerometer first need to be set such that its sample rate is at
least twice superior to the maximum frequency of the vibrations to
be detected.
[0150] Preferably, the accelerometer is highly sensitive. To allow
detection of the vibrations by the accelerometer, the accelerometer
may be bound to the writing tip 6 of the handwriting instrument 2
by rigid contacts with little damping.
[0151] In an embodiment, it is possible to enhance the precision of
the vibration detection by using a support presenting a rough
surface with known spatial frequency.
[0152] In FIG. 10A, representing the collect phase, the
accelerometer is set with a sample rate F2. While the user is using
the handwriting instrument 2, the accelerometer acquires data at
step S20. These data can be sent by short-range radio to a
recording device at step S21.
[0153] In an embodiment, during the collect phase, if the
handwriting instrument 2 also comprises a three-axis gyroscope as
another motion sensor 7, the three-axis gyroscope can also acquire
data that are sent to the recording device at step S21.
[0154] FIG. 10B illustrates the training phase of the neural
network.
[0155] At step S22, the data sent to the recording device are
provided. The data are analyzed at step S23A to determine the
labels (step S238). For example, the labels comprise the strokes
timestamps, detected when vibration is detected in the data, and
the stroke velocity. The stroke velocity is advantageously
determined using the acceleration data and the high frequencies
contained in the vibration.
[0156] Step S24 comprises the undersampling of the data.
Particularly, during the preceding steps, the frequency of the
accelerometer was set to be higher than the one set for the
inference phase. Moreover, the vibration analysis was made on the
basis of the three-axis accelerometer and the three-axis gyroscope.
However, the constant use of the gyroscope leads to high energy
consumption.
[0157] The undersampling step S24 comprises the degradation of the
parameters. Frequency F2 of the accelerometer is reduced to a
frequency F1, smaller than F2, and the training is made only
according to three-axis detection.
[0158] At step S25, the neural network is trained to be able to
perform strokes segmentation, as described with reference to FIG.
8, step S2.
[0159] FIG. 10C illustrates the inference phase. In this phase, the
neural network is trained to detect handwriting problems by means
of strokes segmentation.
[0160] At step S26, a user is using the handwriting instrument 2 in
view of detecting an eventual handwriting problem.
[0161] The accelerometer in the handwriting instrument is set to
the frequency F1 and advantageously, the data are acquired
according to three-axis.
[0162] At step S27, the trained neural network is feed with the
acquired data, At step S28, the neural network is able to deliver
the strokes timestamps and the velocity.
[0163] Finally, the neural network is able to perform the
intermediate stroke feature extraction and the classification at
step S29. Step S29 actually corresponds to steps S4 to S7, already
described with reference to FIG. 8.
[0164] In an embodiment, the neural network can be trained
continuously with the data acquired by the user of the handwriting
pen 2 after the storage of the neural network.
[0165] In an embodiment, the neural network can also be trained to
detect a wrong ductus of the user. The ductus corresponds to the
formation of letter and number.
[0166] More specifically, the neural network is able to determine
if a sequence of strokes correspond to a letter or a number.
[0167] To this end, the neural network can also be fed with a large
data base of letters and numbers, Each letters and numbers can be
associated with a sequence of strokes. The sequence of strokes can
advantageously corresponds to acceleration signals acquired by the
accelerometer during the collect phase when forming the letters and
numbers.
[0168] The labels to be determined by the neural network may be the
direction and an order of the sequence of strokes for each letter
and number.
[0169] In step S5 of FIG. 8, the intermediate features can then
also comprise the temporal sequence of strokes and their
direction.
[0170] In step S7, the neural network is able to determine if the
user is forming correctly letters and numbers.
* * * * *