U.S. patent application number 17/548569 was filed with the patent office on 2022-06-02 for system and method of real-time video overlaying or superimposing display.
This patent application is currently assigned to LOOP NOW TECHNOLOGIES INC.. The applicant listed for this patent is LOOP NOW TECHNOLOGIES INC.. Invention is credited to Li CHEN, Yang-Cheng FAN, WuHsi LI, Ting Kwan LUK, Xinshuo ZHANG.
Application Number | 20220174338 17/548569 |
Document ID | / |
Family ID | |
Filed Date | 2022-06-02 |
United States Patent
Application |
20220174338 |
Kind Code |
A1 |
CHEN; Li ; et al. |
June 2, 2022 |
SYSTEM AND METHOD OF REAL-TIME VIDEO OVERLAYING OR SUPERIMPOSING
DISPLAY
Abstract
Technologies relating to system and method of real-time video
overlaying or superimposing display from multiple mutually
synchronous cameras. are disclosed. An example method of real-time
video overlaying includes the steps of: synchronizing frame rates
of a depth data, a face meta data, and a video data of a first
camera video output captured by a first camera; determining a first
depth between a user face and the first camera; using a cutoff
depth to determine a user body contour; generating a binary mask of
the user body contour based on the first depth and the cutoff
depth; smoothing an edge of the binary mask; merging the binary
mask with the first camera video output and generating a merged
first camera video output; and overlaying the merged first camera
video output onto a second camera video output.
Inventors: |
CHEN; Li; (San Jose, CA)
; FAN; Yang-Cheng; (San Jose, CA) ; LUK; Ting
Kwan; (Saratoga, CA) ; ZHANG; Xinshuo; (Falcon
Heights, MN) ; LI; WuHsi; (Cambridge, MA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
LOOP NOW TECHNOLOGIES INC. |
Redwood City |
CA |
US |
|
|
Assignee: |
LOOP NOW TECHNOLOGIES INC.
Redwood City
CA
|
Appl. No.: |
17/548569 |
Filed: |
December 12, 2021 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
16839081 |
Apr 3, 2020 |
11223855 |
|
|
17548569 |
|
|
|
|
62902361 |
Sep 18, 2019 |
|
|
|
International
Class: |
H04N 21/234 20060101
H04N021/234; H04N 5/247 20060101 H04N005/247; H04N 5/272 20060101
H04N005/272; H04N 21/2187 20060101 H04N021/2187 |
Claims
1. A real-time video overlaying system comprising: a user device
comprising a device body having a first side and a second side; a
first camera formed on the first side; a second camera formed on
the second side; a depth sensor formed on the first side; and an
image processor, wherein the image processor comprises: a
synchronizer unit configured to synchronize a first video data
captured by the first camera, a face meta data captured by the
first camera, and a depth data captured by the depth sensor; a
binary mask generator unit configured to generate a binary mask
from the first video data captured by the first camera; a video
merger unit configured to merge the binary mask into the first
video data, and generating a merged first video data; and a video
overlaying unit configured to overlay the merged first video data
onto a second video data captured by the second camera.
2. The system as claimed in claim 1, wherein the user device
comprises a mobile phone or a wearable device.
3. The system as claimed in claim 1, further comprising: an image
sensor; and a memory.
4. The system as claimed in claim 3, wherein the image sensor
comprises a CCD or a CMOS sensor.
5. The system as claimed in claim 1, further comprising a video
smoothing unit configured to smooth an edge of the binary mask.
6. The system as claimed in claim 1, further comprising a video
transforming unit configured to transform the binary mask to allow
a drag-and-zoom feature.
7. The system as claimed in claim 1, wherein the binary mask
generator unit is configured to generate the binary mask of a
user's body contour.
8. The system as claimed in claim 1, wherein the first camera, the
second camera, and the depth sensor are clocked differently.
9. The system as claimed in claim 5, wherein the first camera and
the depth sensor have different frame rates.
10. The system as claimed in claim 1, wherein the depth sensor is
embedded in the first camera.
11. The system as claimed in claim 1, wherein the first video data
is a selfie video.
12. The system as claimed in claim 3, wherein the memory comprises
a hard drive or a flash memory.
Description
CROSS-REFERENCE TO RELATED APPLICATION
[0001] The present application is a divisional application of U.S.
patent application Ser. No. 16/839,081, filed on Apr. 3, 2020,
which claims the benefit of priority of U.S. provisional
application Ser. No. 62/902,361, filed on Sep. 18, 2019, the entire
contents of which are incorporated herein by reference.
TECHNICAL FIELD
[0002] The present disclosure relates generally to system and
method of real-time video overlaying or superimposing display and
more specifically to system and method of real-time video
overlaying or superimposing display from multiple mutually
synchronous cameras.
BACKGROUND
[0003] Traditionally, background overlaying and background
superimposing are image processing techniques used in production of
visual objects such as images, photographs, videos, films, movies,
or motion pictures.
[0004] As more and more video contents are produced via mobile
phones or any other wearable devices. Conventional techniques such
as capturing videos or images from a camera and overlaying the
background in a personal computer thereafter are outdated and
time-consuming. Further than that, because live streaming platform
such as YouTube, Twitch, Mixer, or other real-time video streaming
has a trendy need in creating fashionable video contents, these
live streaming content creators would like to have the overlaying
techniques to be more accessible, flexible, and easier to use. For
instance, if these techniques can be achieved by a system or a
method executed via a mobile phone or a wearable device with
cameras, a live streamer may overlay or superimpose a background
with her selfie in the palm of her hand.
SUMMARY
[0005] Technologies relating to system and method of real-time
video overlaying or superimposing display from multiple mutually
synchronous cameras are disclosed.
[0006] An exemplary real-time video overlaying system, in some
implementations, includes: a user device including a device body
having a first side and a second side; a first camera formed on the
first side; a second camera formed on the second side; a depth
sensor formed on the first side; and an image processor, wherein
the image processor includes: a synchronizer unit configured to
synchronize a first video data captured by the first camera, a face
meta data captured by the first camera, and a depth data captured
by the depth sensor; a binary mask generator unit configured to
generate a binary mask from the first video data captured by the
first camera; a video smoothing unit configured to smooth an edge
of the binary mask; a video transforming unit configured to
transform the binary mask to allow a drag-and-zoom feature; a video
merger unit configured to merge the binary mask into the first
video data, and generating a merged first video data; and a video
overlaying unit configured to overlay the merged first video data
onto a second video data captured by the second camera.
[0007] In some implementations, the user device includes a mobile
phone or a wearable device.
[0008] The real-time video overlaying system, in some
implementations, further includes: an image sensor; and a
memory.
[0009] In some implementations, the image sensor includes a CCD or
a CMOS sensor, and the memory includes a hard drive or a flash
memory.
[0010] An exemplary method of real-time video overlaying includes
the steps of: synchronizing frame rates of a depth data, a face
meta data, and a video data of a first camera video output captured
by a first camera; determining a first depth between a user face
and the first camera; using a cutoff depth to determine a user body
contour; generating a binary mask of the user body contour based on
the first depth and the cutoff depth; smoothing an edge of the
binary mask; merging the binary mask with the first camera video
output and generating a merged first camera video output; and
overlaying the merged first camera video output onto a second
camera video output.
[0011] In some implementations, the first camera and the second
camera are facing in the opposite direction.
[0012] In some implementations, the first depth is determined by a
depth sensor.
[0013] In some implementations, the step of synchronizing frame
rates includes synchronizing the frame rate to 60 FPS.
[0014] In some implementations, any point on the binary mask
further than the user body contour is set a value of 0, and any
image point falls within the user body contour is set to a value of
1.
[0015] The method of real-time video overlaying, in some
implementations, further includes: applying a low pass filter on
the binary mask after generating the binary mask of the user body
contour.
[0016] In some implementations, the step of applying the low pass
filter on the binary mask after generating the binary mask of the
user body contour further includes: temporarily buffering a last
binary mask that corresponds to a last frame, and then mixing the
last binary mask with a current binary mask that corresponds to a
current frame, and then generating the binary mask according to a
combination of the last frame and the current frame.
[0017] In some implementations, the combination of the last frame
and the current frame follows an exemplary equation: the binary
mask=last frame*0.4+current frame*0.6.
[0018] In some implementations, the step of smoothing an edge of
the binary mask further includes: creating a first alpha matte on
the binary mask.
[0019] In some implementations, the step of smoothing the edge of
the binary mask further includes: upscaling the first alpha matte
to RGB resolution after creating the first alpha matte on the
binary mask.
[0020] The method of real-time video overlaying, in some
implementations, further includes a step of: smoothing the binary
mask by applying a gamma adjust process and a gaussian blur
process; and transforming the binary mask to allow a drag-and-zoom
feature.
[0021] In some implementations, the step of applying the gamma
adjust process further includes applying a CIGammaAdjust
filter.
[0022] In some implementations, the step of applying the gaussian
blur process further includes applying a CIGaussianBlur filter.
[0023] In some implementations, the step of transforming the binary
mask further includes: scaling the binary mask to the same size as
the second camera video output.
[0024] In some implementations, the step of transforming the binary
mask further includes: placing the binary mask at a position
corresponding to a bottom center of the second camera video output
after scaling the binary mask to the same size as the second camera
video output.
[0025] In some implementations, the step of merging the binary mask
with the first camera video output and generating the merged first
camera video output further includes: applying a second alpha matte
to the first camera video output; correcting an orientation of the
first camera video output; and applying a CIBlendWithMask filter to
merge the first camera video output with the binary mask.
BRIEF DESCRIPTION OF THE DRAWINGS
[0026] FIG. 1 is a schematic diagram illustrating an example
real-time video overlaying system in accordance with some
implementations of the present disclosure.
[0027] FIG. 2 is a flowchart illustrating a method of real-time
video overlaying and superimposing display from multiple mutually
synchronous cameras in accordance with some implementations of the
present disclosure.
[0028] FIG. 3 is a flowchart illustrating a method of real-time
video overlaying and superimposing display from multiple mutually
synchronous cameras in accordance with some implementations of the
present disclosure.
[0029] FIG. 4A is an exemplary image illustrating the exemplary
extracted image using a DeepLabV3 model.
[0030] FIG. 4B is an exemplary image illustrating the exemplary
extracted image using a TrueDepth 3D sensor.
[0031] FIG. 4C is an exemplary figure map illustrating Atrous
Convolution with different rates r in accordance with DeepLabV3
implementations.
[0032] FIG. 4D is the exemplary equation of DeepLabV3 in accordance
with DeepLabV3 implementations.
[0033] FIG. 5 is a schematic diagram illustrating an exemplary
real-time video overlaying system in accordance with some
implementations of the present disclosure.
[0034] FIG. 6A is an exemplary image illustrating a result of the
real-time video overlaying system and methods in accordance with
some implementations of the present disclosure.
[0035] FIG. 6B is an exemplary image illustrating a result of the
real-time video overlaying system in accordance with some
implementations of the present disclosure.
[0036] The implementations disclosed herein are illustrated by way
of example, and not by way of limitation, in the figures of the
accompanying drawings. Like reference numerals refer to
corresponding parts throughout the drawings.
DETAILED DESCRIPTION
[0037] Technologies relating to system and method of real-time
video overlaying or superimposing display from multiple mutually
synchronous cameras are disclosed. The technologies described in
the present disclosure may provide the following technical
advantages.
[0038] First, the disclosed technology utilizes only a mobile
device or a wearable device to perform real-time video overlaying
display from two synchronous cameras in one device which is highly
favorable for live online streaming applications.
[0039] Second, with a synchronizing process applying on video data,
depth data, and face meta data captured by the front camera, the
disclosed technology may generate a binary mask of the same frame
rate with live video and allow real-time video processing. This
enables a subsequent overlaying process to overlay the face video
with the background video.
[0040] Third, the disclosed technology provides several smoothing
and transforming processes on the real-time video which makes the
video fit into the background without looking like a shadow or
silhouette. This also includes techniques that make the live video
streamer to be able to jump or move dynamically. Some techniques of
the present disclosure provide drag-and-zoom functions as well. The
present disclosure allows live online streamers to interact with
audiences much easier and more flexible in any kind of outdoor
video.
[0041] Finally, the disclosed technology may not only overlay or
superimpose the selfie video with background video but also overlay
the selfie video with another selfie video in another camera. This
provides more possibilities and capabilities for live streaming
applications in the future.
[0042] FIG. 1 is a schematic diagram 1000 illustrating an example
real-time video overlaying system 100 in accordance with some
implementations of the present disclosure.
[0043] As shown in FIG. 1, the real-time video overlaying system
100 includes a user device 130 and an image processing unit 120.
The user device 130 includes a device body 131 having a first side
1311 and a second side 1313; a first camera 1331 formed on the
first side 1311 of the device body 131; a second camera 1333 formed
on the second side 1311 of the device body 131; a depth sensor 1335
formed on the first side 1311 of the device body 131; and a display
135 formed on the first side 1311 of the device body 131.
[0044] In some implementations, the user device 130 includes an
electronic device such as a mobile phone, or a wearable device.
[0045] In some implementations, the first camera 1331 is a front
camera of a mobile phone, and the second camera 1333 is a back
camera of the mobile phone. Therefore, the two cameras are formed
at opposite sides of the user device 131 and thus have opposite
camera view. This allows the user device 130 to capture a first
video of a first user face 141 and a second video of a background
view 143 synchronously with contradictory visions.
[0046] In some implementations, the depth sensor 1335 may be
embedded on the first camera 1331.
[0047] Next, the image processing unit 120 includes an image sensor
121, an image processor 123, and a memory 125.
[0048] In some implementations, the image processing unit 120 may
be implemented in the user device 130. In some implementations, the
image processing unit 120 may be a part of the user device 130. For
example, it can be embedded in the device body 131. In some
implementations, a part of the image processing unit 130 may be
remotely located outside of the user device (i.e., the image
processor 123 may be located beyond the user device) because it can
be a cloud-based computation, and therefore, it is not arranged on
the user device 130.
[0049] In some implementations, the image sensor 121 is configured
to convert light waves captured via cameras into signals such as
analog signals, digital signals or optical signals. In some
implementations, the image sensor 121 includes a charge-coupled
device (CCD) and the active-pixel sensor (CMOS sensor).
[0050] In some implementations, the image processor 123 is
configured to provide several image processing units including (1)
a synchronizer unit 1231 configured to synchronize a first video
data captured by the first camera 1331, a face meta data captured
by the first camera 1331, and a depth data captured by the depth
sensor 1335, (2) a binary mask generator unit 1232 configured to
generate a binary mask from the first video data captured by the
first camera 1331 to be processed and modified, (3) a video
smoothing unit 1233 configured to smooth an edge of the binary
mask, (4) a video transforming unit 1234 configured to transform
the binary mask to allow a drag-and-zoom feature, (5) a video
merging unit 1235 configured to merge the binary mask with the
first video data, and generating a merged first video data, and (6)
a video overlaying unit 1236 configured to overlay the merged first
video data onto a second video data captured by the second camera.
The image processing units and their functions of the above image
processor 123 may be implemented by using image processing methods
and corresponding codes or algorithms in accordance with some
implementations of the present disclosure. These image processing
methods and corresponding codes and algorithms will be discussed in
detail in a subsequent section.
[0051] In some implementations, the memory 125 includes a hard
drive or a flash memory. The memory 125 is configured to provide
temporary or permanent data storage including storage of depth
data, image data, or video data before, during, or after the
real-time video overlaying process.
[0052] FIG. 2 is a flowchart 2000 illustrating a method 200 of
real-time video overlaying and superimposing display from multiple
mutually synchronous cameras in accordance with some
implementations of the present disclosure.
[0053] As shown in FIG. 2, the method 200 includes capturing a
first video via a front camera 201; and synchronizing a face meta
data 205, a depth data 207, and the first video of the front camera
209 into the same frame rate, e.g., a 60 frames-per-second (FPS).
As mentioned above, the front camera, the back camera, and the
depth sensor may be clocked differently. For example, the front
camera and the depth sensor may have different frame rates. For
them to work together, a synchronizing process is required to
synchronize the frame rate.
[0054] In some implementations, an AVCaptureDataOutputSynchronizer
from Apple may be used to achieve the synchronizing process. The
AVCaptureDataOutputSynchronizer is an object that coordinates
time-matched delivery of data from multiple capture outputs.
[0055] Next, after the synchronizing process, a real-time body
segmentation process and an edge-smoothing process are
performed.
[0056] Conventionally, a segmentation process may be done by
processing the front camera output frame by frame by using the
DeepLabV3 model or TrueDepth 3D sensor. FIG. 4A is an image
illustrating the exemplary extracted image using the DeepLabV3
model. FIG. 4B is an image illustrating the exemplary extracted
image using the TrueDepth 3D sensor.
[0057] FIG. 4C illustrates the exemplary figure map for Atrous
Convolution with different rates r in accordance with DeepLabV3
implementations. And FIG. 4D illustrates the exemplary equation of
DeepLabV3 in accordance with DeepLabV3 implementations. For
DeepLabV3, each location I on the output y and a filter w, the
Atrous Convolution is applied over the input feature map x where
the astrous rate r corresponds to the stride with which we sample
the input signal. This is equivalent to convolving the input x with
upsampled filters produced by inserting (r-1) zeros between two
consecutive filter values along each spatial dimension, (trous
means holes in English.) When r=1, it is standard convolution. By
adjusting r, we can adaptively modify the filter's field-of-view.
It is also called dilated convolution (DilatedNet) or Hole
Algorithm.
[0058] Top: Standard convolution
[0059] Bottom: Atrous convolution
[0060] We can see that when r=2, the input signal is sampled
alternatively. First, pad-2 means we pad 2 zeros at both the left
and right sides. Then, with r=2, we sample the input signal every 2
inputs for convolution. Atrous convolution allows us to enlarge the
field of view of filters to incorporate large context. It thus
offers an efficient mechanism to control the field-of-view and
finds the best trade-off between accurate localization (small
field-of-view) and context assimilation (large field-of-view). In
some examples, Atrous convolution applies a mechanism named Atrous
Spatial Pyramid Pooling (ASPP). However, these two approaches may
not deliver results in real-time.
[0061] Back to implementations of the present disclosure, the depth
data hereinafter is important to aid the body segmentation in the
present disclosure. And some smoothing techniques are also used to
make sure the results are not degraded.
[0062] As shown in FIG. 2, the method 200 further includes checking
whether there is a face in the face meta data feed 205. If so,
determining a center point of the face, and collecting a depth data
207 between the center point of the face and the front camera. A
cutoff depth is determined in advance to allow the whole body to
appear in the video and to account for distance measuring errors.
In some implementations, the cutoff depth is 0.25 meters.
[0063] In some implementations, the method 200 further includes
setting a maximum depth, so that the user can do something fun like
"jumping" into the video from nowhere. Therefore, in some
implementations, the maximum depth may be set to 1.2 meters to
allow for the "jump-in" effect of the video.
[0064] The exemplary codes to execute the steps above of the method
200 may be as follows:
TABLE-US-00001 if let firstFace =
syncedMetaData?.metadataObjects.first, let connection =
sessionManager.frontCameraVideoDataOutput.connection(with:
AVMediaType.video), let face =
sessionManager.frontCameraVideoDataOutput.transformedMetadataObject(for:
firstFace, connection: connection) { let faceCenter = CGPoint(x:
face.bounds.midX, y: face.bounds.midY) let scaleFactor =
CGFloat(CVPixelBufferGetWidth(depthPixelBuffer)) /
CGFloat(CVPixelBufferGetWidth(videoPixelBuffer)) let pixelX =
Int((faceCenter.x * scaleFactor).rounded( )) let pixelY =
Int((faceCenter.y * scaleFactor).rounded( )) if pixelX <
depthHeight && pixelX >= 0 && pixelY <
depthWidth && pixelY >= 0 {
CVPixelBufferLockBaseAddress(depthPixelBuffer, .readOnly) let
rowData = CVPixelBufferGetBaseAddress(depthPixelBuffer)! + pixelX *
CVPixelBufferGetBytesPerRow(depthPixelBuffer) let faceCenterDepth =
rowData.assumingMemoryBound(to: Float32.self)[pixelY]
CVPixelBufferUnlockBaseAddress(depthPixelBuffer, .readOnly) // 1.2
is a number to allow for "jump-in" effect
sessionManager.depthCutOff = min(faceCenterDepth + 0.25, 1.2) }
}
[0065] Next, in order to perform real-time video processing, the
method 200 further includes generating a binary mask 213 of the
user's body contour by using the cutoff depth from the previous
step. Any point further than the user's body contour will be set a
value of 0 and will eventually be filtered off. Also, with the aid
of the binary mask, any image point falls within the user's body
contour will be assigned a non-zero value, e.g., 1, such that the
image point will be kept. By assigning binary values to image
points, the method 200 may effectively process image points without
introducing a heavy cost. This enables the system to do the
calculation in real-time at, for example, 60 FPS.
[0066] Although it is a great solution to do real-time video
processing, it should be noted that the depth data feed has a very
low resolution. Therefore, the resulting image is very sensitive to
lighting conditions, especially on the edges of the body contour.
Moreover, such a phenomenon will render a constantly changing
binary mask which may sabotage the final result.
[0067] To deal with such a phenomenon, the method 200 further
includes applying a low pass filter (LPF) 215 on the binary mask
213. More specifically, the method 200 further includes temporarily
buffering a last binary mask that corresponds to a last frame
(lastData), and then mixing the last binary mask with a current
binary mask that corresponds to a current frame (currentData). In
this way, a final result is generated according to the following
exemplary equation:
filteredResult=lastData*0.4+currentData*0.6
[0068] With the aid of the LPF, this generates a good final result
at a low computational cost.
[0069] One of the exemplary codes to achieve the above goal may be
as follows:
TABLE-US-00002 CVPixelBufferLockBaseAddress(depthPixelBuffer,
CVPixelBufferLockFlags(rawValue: 0)) if let
lastDepthDataPixelBuffer = lastDepthDataPixelBuffer {
CVPixelBufferLockBaseAddress(lastDepthDataPixelBuffer,
CVPixelBufferLockFlags(rawValue: 0)) if
CVPixelBufferGetWidth(lastDepthDataPixelBuffer) != depthWidth ||
CVPixelBufferGetHeight(lastDepthDataPixelBuffer) != depthHeight {
self.lastDepthDataPixelBuffer = nil } } for yMap in 0 ..<
depthHeight { let rowData =
CVPixelBufferGetBaseAddress(depthPixelBuffer)! + yMap *
CVPixelBufferGetBytesPerRow(depthPixelBuffer) let lastRowData :
UnsafeMutableRawPointer? = (lastDepthDataPixelBuffer != nil) ?
CVPixelBufferGetBaseAddress(lastDepthDataPixelBuffer!)! + yMap *
CVPixelBufferGetBytesPerRow(lastDepthDataPixelBuffer!) : nil let
data = UnsafeMutableBufferPointer<Float32>(start:
rowData.assumingMemoryBound(to: Float32.self), count: depthWidth)
let lastData : UnsafeMutableBufferPointer<Float32>? =
(lastRowData != nil) ?
UnsafeMutableBufferPointer<Float32>(start:
lastRowData!.assumingMemoryBound(to: Float32.self), count:
depthWidth) : nil for index in 0 ..< depthWidth { if data[index]
> 0 && data[index] <= sessionManager.depthCutOff {
data[index] = 1.0 }else { data[index] = 0.0 } if let lastData =
lastData { let lowPassFilterFactor:CGFloat = 0.4 let current =
data[index] * Float32(1.0-lowPassFilterFactor) let last =
lastData[index] * Float32(lowPassFilterFactor) data[index] =
current + last } } } if let lastDepthDataPixelBuffer =
lastDepthDataPixelBuffer {
CVPixelBufferUnlockBaseAddress(lastDepthDataPixelBuffer,
CVPixelBufferLockFlags(rawValue: 0)) }
CVPixelBufferUnlockBaseAddress(depthPixelBuffer,
CVPixelBufferLockFlags(rawValue: 0)) lastDepthDataPixelBuffer =
depthPixelBuffer.copy( )
[0070] Because the binary mask 213 may still introduce pixelated
edges and, under an outdoor environment with strong lighting, the
user's body contour may be partially cut off due to the roughness
of the depth sensing, the method 200 further includes a smoothing
process to generate a smoothed mask 219. First, the method 200
includes smoothing an edge of the binary mask by creating a first
alpha matte on the binary mask. Second, the method 200 includes
smoothing the binary mask by applying a gamma adjust process and a
gaussian blur process. Third, the method 200 further includes
transforming the binary mask to allow a drag-and-zoom feature. It
is noted that alpha matte is opaque when a pixel value of the alpha
channel is 100%, where the alpha channel is used for communicating
transparency information.
[0071] In some implementations, the step of applying the gamma
adjust process includes applying a CIGammaAdjust filter.
[0072] In some implementations, the step of applying the gaussian
blur process includes applying a CIGaussianBlur filter.
[0073] One of the exemplary codes to achieve the above steps may be
as follows:
TABLE-US-00003 // Create the mask from that pixel buffer. let
depthOrientation =
Util.imageOrientationToFixCaptureVideoOrientation(videoOrientation:
sessionManager.depthDataOutputConnection!.videoOrientation) let
depthMaskImage = CIImage(cvPixelBuffer: depthPixelBuffer, options:
[:]).oriented(depthOrientation) let (blurRadius, gamma) =
Util.superimposeConfig(brightness: lastBrightness) // Smooth edges
to create an alpha matte, then upscale it to the RGB resolution. if
frontVideoWidth == nil { frontVideoWidth =
CVPixelBufferGetWidth(videoPixelBuffer) } if frontVideoHeight ==
nil { frontVideoHeight = CVPixelBufferGetHeight(videoPixelBuffer) }
// Used max to be orientation-safe let alphaUpscaleFactor =
Float(max(frontVideoHeight!, frontVideoWidth!)) /
Float(max(depthWidth, depthHeight)) if frontImageTransform == nil {
let backVideoWidth = backCamImage.extent.size.width let
backVideoHeight = backCamImage.extent.size.height // Place front
cam video at the bottom center let scale = backVideoWidth /
CGFloat(frontVideoWidth!) frontImageTransform =
CGAffineTransform(scaleX: scale, y: scale) .translatedBy( x: 0.0 /
scale * backVideoWidth, y: 0.25 / scale * backVideoHeight)
lastFrontImageTransform = frontImageTransform! } var blurredMask =
depthMaskImage.clampedToExtent( ) .applyingFilter("CIGaussianBlur",
parameters: ["inputRadius": blurRadius])
.applyingFilter("CIGammaAdjust", parameters: ["inputPower": gamma])
.cropped(to: depthMaskImage.extent)
[0074] Fourth, in order for the filter combination to work great in
different lighting conditions, a brightness data from Exchangeable
image file format (EXIF) data of each frame is extracted, then
using the brightness data to derive different filter parameter
combinations. It is noted that EXIF is a standard that specifies
the formats for images, sound, and ancillary tags used by digital
cameras (including smartphones), scanners and other systems
handling image and sound files recorded by digital cameras.
[0075] One of the exemplary codes to achieve the above step may be
as follows:
TABLE-US-00004 public static func
getBrightness(sampleBuffer:CMSampleBuffer) -> Double {
//Retrieving EXIF data of camera frame buffer let rawMetadata =
CMCopyDictionaryOfAttachments(allocator: nil, target: sampleBuffer,
attachmentMode:
CMAttachmentMode(kCMAttachmentMode_ShouldPropagate)) let metadata =
CFDictionaryCreateMutableCopy(nil, 0, rawMetadata) as
NSMutableDictionary let exifData = metadata.value(forKey: "{Exif}")
as? NSMutableDictionary let brightness = exifData?.object(forKey:
kCGImagePropertyExifBrightnessValue) as? Double return brightness
?? 0.0 } // (blurRadius, gamma) public static func
superimposeConfig(brightness:Double) -> (Float, Float) { if
brightness >= 7.8 { return (10.0, 0.12) } else if brightness
> 5.0 { return (5.0, 0.2) } else { return (1.0, 1.0) } }
[0076] Next, the method 200 includes merging the smoothed binary
mask 219 with a first camera video output 209 which may be a selfie
video output. A smoothed selfie video output 221 may be generated
thereafter. To be more specific, in some implementations, the
method 200 includes applying a second alpha matte to the selfie
video output, correcting an orientation of the selfie video output,
and applying CIBlendWithMask Filter to blend the selfie video
output with the smoothed binary mask. A smoothed selfie video 221
is generated thereafter.
[0077] One of the exemplary codes to achieve the above steps may be
as follows:
TABLE-US-00005 let alphaMatte = blurredMask
.applyingFilter("CIBicubicScaleTransform", parameters:
["inputScale": alphaUpscaleFactor]) .transformed(by:
frontImageTransform!) let frontCamImageOrientation =
Util.imageOrientationToFixCaptureVideoOrientation(videoOrientation:
sessionManager.frontCameraVideoDataOutputConnection!.videoOrientation)
var frontCamImage = CIImage(cvPixelBuffer:
videoPixelBuffer).oriented(frontCamImageOrientation) frontCamImage
= frontCamImage.transformed(by: frontImageTransform!) // Apply
alpha matte to the video. let parameters = [ "inputMaskImage":
alphaMatte, "inputBackgroundImage": backCamImage ] let mergedImage
= frontCamImage.applyingFilter("CIBlendWithMask", parameters:
parameters)
[0078] Finally, the method includes overlaying the smoothed selfie
video 221 onto the back camera video 233 captured by the back
camera. An overlaid video 235 is generated as a result.
[0079] FIG. 3 is a flowchart 3000 illustrating a method 300 in
accordance with some implementations of the present disclosure.
[0080] As shown in FIG. 3, the method 300 includes synchronizing
frame rates of a depth data, a face meta data, and a video data of
a first camera video output captured by a first camera (step 302);
determining a first depth between a user face and the first camera
via a depth sensor, and using a cutoff depth to determine a user
body contour (step 304); generating a binary mask of the user body
contour based on the first depth and the cutoff depth (step 306);
applying a low pass filter on the binary mask (step 308); smoothing
an edge of the binary mask by creating a first alpha matte onto the
edge of the binary mask (step 310); smoothing the binary mask by
applying a gamma adjust process and a gaussian blur process (step
312); transforming the binary mask to allow a drag-and-zoom feature
(step 314); merging the binary mask with the first camera video
output and generating a merged first camera video output (step
316); and overlaying the merged first camera video output onto a
second camera video output captured by a second camera (318). In
some implementations, the method 300 may be specified by adding
more detailed steps of the method 200.
[0081] In some implementations, the first camera and the second
camera are facing in the opposite direction.
[0082] In some implementations, the step of transforming the binary
mask includes scaling the binary mask to the same size (height and
width) as the second camera video output.
[0083] In some implementations, the step of transforming the binary
mask further includes placing the binary mask at a position
corresponding to a bottom center of the second camera video
output.
[0084] In some implementations, the step of smoothing an edge of
the binary mask by creating a first alpha matte onto the edge of
the binary mask further includes upscaling the first alpha matte to
RGB resolution.
[0085] FIG. 5 is a schematic diagram 5000 illustrating an exemplary
real-time video overlaying system 500 in accordance with some
implementations of the present disclosure.
[0086] As shown in FIG. 5, it is basically the same as in FIG. 1.
However, the real-time video overlaying system 100 is configured to
overlay a selfie video of a front camera to a background video of a
back camera, while the real-time video overlaying system 500 is
configured to overlay a first selfie video of a front camera 5331
to a second selfie video of a back camera. Since based on the
method 200 and 300, coordinates of the front user face contour and
the back user face contour may be detected and determined by the
front and back camera accordingly, by overlaying and superimposing
the front user face contour to the back user face contour according
to the coordinates, the back user face may be replaced by the front
user face. Furthermore, with the aid of the real-time coordinated
and boundaries of the back user face contour, the front user face
contour may be dynamically resized and placed over the back user
face contour. FIG. 6A is an exemplary image illustrating a result
of the real-time video overlaying system 100 and the method 200,
300 in accordance with some implementations of the present
disclosure. FIG. 6B is an exemplary image illustrating a result of
the real-time video overlaying system 500 in accordance with some
implementations of the present disclosure.
[0087] Plural instances may be provided for components, operations
or structures described herein as a single instance. Finally,
boundaries between various components, operations, and data stores
are somewhat arbitrary, and particular operations are illustrated
in the context of specific illustrative configurations. Other
allocations of functionality are envisioned and may fall within the
scope of the implementation(s). In general, structures and
functionality presented as separate components in the example
configurations may be implemented as a combined structure or
component. Similarly, structures and functionality presented as a
single component may be implemented as separate components. These
and other variations, modifications, additions, and improvements
fall within the scope of the implementation(s).
[0088] It will also be understood that, although the terms "first,"
"second," etc. may be used herein to describe various elements,
these elements should not be limited by these terms. These terms
are only used to distinguish one element from another. For example,
a first column could be termed a second column, and, similarly, a
second column could be termed the first column, without changing
the meaning of the description, so long as all occurrences of the
"first column" are renamed consistently and all occurrences of the
"second column" are renamed consistently. The first column and the
second are columns both column s, but they are not the same
column.
[0089] The terminology used herein is for the purpose of describing
particular implementations only and is not intended to be limiting
of the claims. As used in the description of the implementations
and the appended claims, the singular forms "a", "an" and "the" are
intended to include the plural forms as well, unless the context
clearly indicates otherwise. It will also be understood that the
term "and/or" as used herein refers to and encompasses any and all
possible combinations of one or more of the associated listed
items. It will be further understood that the terms "comprises"
and/or "comprising," when used in this specification, specify the
presence of stated features, integers, steps, operations, elements,
and/or components, but do not preclude the presence or addition of
one or more other features, integers, steps, operations, elements,
components, and/or groups thereof.
[0090] As used herein, the term "if" may be construed to mean
"when" or "upon" or "in response to determining" or "in accordance
with a determination" or "in response to detecting," that a stated
condition precedent is true, depending on the context. Similarly,
the phrase "if it is determined (that a stated condition precedent
is true)" or "if (a stated condition precedent is true)" or "when
(a stated condition precedent is true)" may be construed to mean
"upon determining" or "in response to determining" or "in
accordance with a determination" or "upon detecting" or "in
response to detecting" that the stated condition precedent is true,
depending on the context.
[0091] The foregoing description included example systems, methods,
techniques, instruction sequences, and computing machine program
products that embody illustrative implementations. For purposes of
explanation, numerous specific details were set forth in order to
provide an understanding of various implementations of the
inventive subject matter. It will be evident, however, to those
skilled in the art that implementations of the inventive subject
matter may be practiced without these specific details. In general,
well-known instruction instances, protocols, structures, and
techniques have not been shown in detail.
[0092] The foregoing description, for purpose of explanation, has
been described with reference to specific implementations. However,
the illustrative discussions above are not intended to be
exhaustive or to limit the implementations to the precise forms
disclosed. Many modifications and variations are possible in view
of the above teachings. The implementations were chosen and
described in order to best explain the principles and their
practical applications, to thereby enable others skilled in the art
to best utilize the implementations and various implementations
with various modifications as are suited to the particular use
contemplated.
* * * * *