CN110751056A - Pedestrian motion prediction method based on improved top-down method multi-person posture detection - Google Patents
Pedestrian motion prediction method based on improved top-down method multi-person posture detection Download PDFInfo
- Publication number
- CN110751056A CN110751056A CN201910921085.XA CN201910921085A CN110751056A CN 110751056 A CN110751056 A CN 110751056A CN 201910921085 A CN201910921085 A CN 201910921085A CN 110751056 A CN110751056 A CN 110751056A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- frame
- bone point
- bone
- posture detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
- G06V40/25—Recognition of walking or running movements, e.g. gait recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a pedestrian motion prediction method based on improved top-down method multi-person posture detection, which extracts a human body posture block diagram, bone points and postures by processing images processed by a spatial transformation network, improved top-down method single-person posture detection and an inverse spatial transformation network, predicts and outputs the next actions of pedestrians by carrying out optical flow processing and training of a long-term and short-term memory neural network on the bone points and postures of the human body.
Description
Technical Field
The invention belongs to the technical field of image processing, relates to a multi-person posture detection method applied to an automatic assistant driving system, and particularly relates to a method for detecting and predicting pedestrian movement by adopting improved multi-person postures, which improves the accuracy and the real-time performance of posture detection under the condition of pedestrian coincidence.
Background
The pedestrian detection technology is widely applied to the field of advanced assistant driving, in the field, the pedestrian detection is often good in environmental conditions, the detection effect is obvious under the condition that pedestrians are not overlapped, the pedestrian detection under the complex conditions is always important research content, and an algorithm for multi-person posture detection based on an improved top-down method is provided and applied to real-time human motion prediction.
Currently, there are two mainstream algorithms for multi-person attitude estimation: a top-down method (Two-step frame) of detecting each human detection frame in an environment first and then independently detecting the pose of each human boundary, and a bottom-based method (Part-based frame) of extremely depending on the pose detection accuracy and also repeatedly estimating the bounding box of a single person due to redundant detection frames; the latter method is to detect all the body nodes in the environment first, then to splice to get the skeleton of many people, but because this method depends on the body nodes of people, when two people are very close to each other, the wrong connection is very easy to happen.
Conventional position checking (SPPE) is highly susceptible to false bounding boxes, and redundant bounding boxes can produce redundant positions, and although most advanced pedestrian recognizers have shown good performance, small errors in positioning and cognition are inevitable, and these errors can cause errors in position checking, especially in methods that rely solely on human detection results, and cannot meet the requirements of current assisted driving systems.
Disclosure of Invention
The invention aims to solve the problem that the existing gesture recognition is inaccurate in detection under the condition that multiple persons are overlapped, and provides a pedestrian motion prediction method based on improved top-down method multiple person gesture detection, so that the gesture estimation can be conveniently carried out under the condition that a human body boundary frame is inaccurate.
The technical scheme adopted by the invention is as follows: a pedestrian motion prediction method based on improved top-down method multi-person posture detection is characterized by comprising the following steps:
step 1: inputting an original multi-person pedestrian video;
step 2: performing pedestrian boundary box SSD processing on the input video to obtain a pedestrian boundary box b;
and step 3: performing space network transformation on the pedestrian boundary frame b obtained in the step 2, and extracting a high-quality human body region frame;
and 4, step 4: carrying out single posture detection on each high-quality human body region frame to obtain a redundant bone point confidence coefficient E;
and 5: carrying out redundancy elimination treatment on the bone points E with redundancy;
step 6: mapping the human body area frame obtained in the step 3 to an original image coordinate to obtain a high-quality area frame in the original image coordinate;
and 7: processing the current frame and the previous N frames in the steps 1 to 4 to respectively obtain bone points E (d) of N +1 picturesi) I 1, 2,., N +1, and drawing the motion track of each bone point for the N +1 bone points;
and 8: performing optical flow processing on the N +1 pictures to obtain a displacement vector epsilon (d) of each bone point;
and step 9: the bone point d of each framejPredicting the new bone point v connection of the next frame to obtain a human skeleton diagram;
step 10: n +1 consecutive frames E (d) obtained in step 7i) Andthe bone point displacement offset epsilon (d) after the optical flow processing in the step 8 is transmitted to a long-short term memory neural network LSTM for training a model;
step 11: generating a skeleton block diagram E every M times of training in the step 10fSaid E isfNamely the real-time pedestrian motion prediction.
The pedestrian posture detection method has the advantages of high pedestrian posture detection rate and strong adaptability, and can predict the action of the pedestrian after the preset time (a large number of experiments prove that the maximum undistorted result is the predicted time of 0.5 second), and the specific expression is as follows:
1) the invention uses the improved multi-person posture detection to improve the accuracy of the pedestrian posture detection under the condition of complicated overlapping and overcome the problem of poor accuracy of the traditional multi-person posture detection.
2) The pedestrian frame diagram is subjected to one redundant screening, the pedestrian bone point diagram is also subjected to one redundant screening, and the calculation time is within an allowable range.
3) The invention predicts the action of the pedestrian in seconds after the preset time on the high-precision multi-person posture detection and has high accuracy.
Drawings
FIG. 1 is an algorithmic flow diagram of an embodiment of the present invention;
FIG. 2 is a schematic diagram of a spatial transform network processing procedure according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating a result of single person gesture detection processing in an embodiment of the present invention;
FIG. 4 is a schematic diagram of an inverse spatial transform network processing procedure according to an embodiment of the present invention;
FIG. 5 is a diagram illustrating the result of streamer processing in an embodiment of the invention;
FIG. 6 is a diagram illustrating a predicted posture of a pedestrian 0.5 seconds later in an embodiment of the present invention.
Detailed Description
In order to facilitate the understanding and implementation of the present invention for those of ordinary skill in the art, the present invention is further described in detail with reference to the accompanying drawings and examples, it is to be understood that the embodiments described herein are merely illustrative and explanatory of the present invention and are not restrictive thereof.
Referring to fig. 1, the method for predicting pedestrian movement based on the improved top-down method multi-person posture detection provided by the invention comprises the following steps:
step 1: inputting an original multi-person pedestrian video.
Step 2: performing pedestrian boundary box SSD processing on the input video to obtain a less accurate pedestrian boundary box b, wherein the processing mode is as follows:
b=(bcx,bcy,bw,bh)=(dwlcx+dcx,dylcy+dcy,dwexp(lw),dhexp(lh))
and is
Where d denotes the position of the prior box, l denotes the predicted position of the bounding box, icx,icyRespectively representing the abscissa and ordinate of the center of the bounding box i, iw,ihWhich respectively represent the width and height of the bounding box i, i can be b, d, l.
And step 3: as shown in FIG. 2, a high-quality human body region frame is extracted from the less accurate pedestrian boundary frame b by using the Space Transformation Network (STN)The treatment method comprises the following steps:
wherein, theta1,θ2,θ3All reflect the human body region frameThe vector coefficients of the coordinate relationship before and after transformation,the coordinates of the region box after the transformation for the spatial transformation network STN.
And 4, step 4: as shown in fig. 3, for each high quality body region boxAdopting CNN single posture detection (SPPE) to obtain a confidence level E of a bone point with redundancy, wherein the higher the confidence level is, the more likely it is a correct human bone point, and the processing mode is as follows:
wherein d isj1、dj2Respectively the position of two bone points, LcIs a line segment composed of two bone points, u is the intermediate coefficient for calculating the integral, and u belongs to [0, 1 ]]P (u) is two bone points dj1、dj2The calculation method of the interpolation is as follows:
p(u)=(1-u)dj1+udj2
and 5: eliminating redundant impurities from bone points E with redundant impurities, and selecting the bone point E with the maximum confidence coefficientmaxFor reference, a threshold is defined η -90% as a criterion for which bone points that are relatively close and similar are eliminated (d)i,dj) And then:
if E (d)i,dj) An output of 1 indicates a bone point diIs redundantMiscellaneous, should be eliminated; if E (d)i,dj) An output of 0 indicates a bone point djAre redundant and should be eliminated.
Step 6, as shown in FIG. 4, frame the human body regionMapping to original image coordinate, namely processing the original image coordinate by using inverse space transformation network (STDN) to obtain high-quality area frame in the original image coordinateThe treatment method is as follows:
wherein
[γ1γ2]=[θ1θ2]-1
γ3=-1×[γ1γ2]θ3
Wherein [ theta ]1θ2]The calculation method comprises the following steps:
θ3the calculation method comprises the following steps:
wherein, W represents a matrix formed by each dimension of the input layer and the output layer of the inverse space transformation network, and J (W, b) represents the position of the pedestrian boundary box b in the inverse space transformation network.
And 7: processing the continuous 5 frames of pictures (the current frame and the previous 4 frames) in the steps 1 to 4 to respectively obtain the bone points E (d) of the 5 picturesi) And i is 1, 2, 3, 4 and 5, and the motion track of each bone point is drawn for the 5 bone points.
And 8: as shown in fig. 5, the 5 pictures are subjected to optical flow processing to obtain displacement vectors ∈ (d) of each bone point, and the processing method is as follows:
v=u+d=[ux+dxuy+dy]T
wherein v is the new position of the bone point in the next frame, u represents the position of the bone point, (x, y) represents the coordinates of the bone point, and (u) represents the coordinates of the bone pointx,uy) Coordinates representing the bone point of the next frame, dx、dyRespectively representing the distance between the bone point and the next frame bone point, I (x, y) representing the pedestrian boundary box where the current bone point is located, J (x + d)x,y+dy) A pedestrian boundary box, w, representing the location of the next frame bone pointx、wyTo define two constants of the window for integration of the optical flow method, the size of which determines the time complexity and the effect of the algorithm, in general, wx、wyThe smaller the effect, the better, but the greater the temporal complexity.
And step 9: the bone point d of each framejAnd predicting the new bone point v connection of the next frame to obtain a human skeleton diagram.
Step 10: five consecutive frames E (d) of step 7i) And step 8, transmitting the bone point displacement offset epsilon (d) after the optical flow processing to a long-short term memory neural network (LSTM) for training a model.
Step 11: as shown in FIG. 6, a skeleton map E is generated every 6 times of training in step 9, i.e. every 30 frames 0.5 seconds in advancefSaid E isfNamely the real-time pedestrian motion prediction.
It should be understood that parts of the specification not set forth in detail are prior art; the above description of the preferred embodiments is intended to be illustrative, and not to be construed as limiting the scope of the invention, which is defined by the appended claims, and all changes and modifications that fall within the metes and bounds of the claims, or equivalences of such metes and bounds are therefore intended to be embraced by the appended claims.
Claims (7)
1. A pedestrian motion prediction method based on improved top-down method multi-person posture detection is characterized by comprising the following steps:
step 1: inputting an original multi-person pedestrian video;
step 2: performing pedestrian boundary box SSD processing on the input video to obtain a pedestrian boundary box b;
and step 3: performing space network transformation on the pedestrian boundary frame b obtained in the step 2, and extracting a high-quality human body region frame;
and 4, step 4: carrying out single posture detection on each high-quality human body region frame to obtain a redundant bone point confidence coefficient E;
and 5: carrying out redundancy elimination treatment on the bone points E with redundancy;
step 6: mapping the human body area frame obtained in the step 3 to an original image coordinate to obtain a high-quality area frame in the original image coordinate;
and 7: processing the current frame and the previous N frames in the steps 1 to 4 to respectively obtain bone points E (d) of N +1 picturesi) I 1, 2,., N +1, and drawing the motion track of each bone point for the N +1 bone points;
and 8: performing optical flow processing on the N +1 pictures to obtain a displacement vector epsilon (d) of each bone point;
and step 9: the bone point d of each framejPredicting the new bone point v connection of the next frame to obtain a human skeleton diagram;
step 10: n +1 consecutive frames E (d) obtained in step 7i) And step 8, the bone point displacement offset epsilon (d) after the optical flow processing is transmitted to a long-short term memory neural network LSTM for training a model;
step 11: generating a skeleton block diagram E every M times of training in the step 10fSaid E isfNamely the real-time pedestrian motion prediction.
2. The method for predicting the pedestrian motion based on the multi-person posture detection of the improved top-down method according to claim 1, wherein the concrete implementation process of the step 2 is as follows:
assuming that a pedestrian boundary box is obtained as b, then:
b=(bcx,bcy,bw,bh)=(dwlcx+dcx,dhlcy+dcy,dwexp(lw),dhexp(lh))
and is
Where d denotes the position of the prior box, l denotes the predicted position of the bounding box, icx,icyRespectively representing the abscissa and ordinate of the center of the bounding box i, iw,ihWhich respectively represent the width and height of the bounding box i, i can be b, d, l.
3. The method for predicting the pedestrian motion based on the multi-person posture detection of the improved top-down method according to claim 1, wherein the concrete implementation process of the step 3 is as follows: extracting a high-quality human body region frame from the pedestrian boundary frame b by adopting a space transformation network STN
4. The method for predicting the pedestrian motion based on the multi-person posture detection of the improved top-down method according to claim 1, wherein the specific implementation process of the step 4 is as follows: for each high quality body region frameAdopting CNN single posture to detect SPPE to obtain a redundant bone point confidence E;
wherein d isj1、dj2Respectively the position of two bone points, LcIs a line segment composed of two bone points, u is the intermediate coefficient for calculating the integral, and u belongs to [0, 1 ]]P (u) is two bone points dj1、dj2The calculation method of the interpolation is as follows:
p(u)=(1-u)dj1+udj2。
5. the method for predicting the pedestrian motion based on the multi-person posture detection of the improved top-down method according to claim 1, wherein the concrete implementation process of the step 5 is as follows: selecting the bone point E with the maximum confidencemaxFor reference, defining η as a standard threshold, then:
if E (d)i,dj) An output of 1 indicates a bone point diAre redundant and should be eliminated; if E (d)i,dj) An output of 0 indicates a bone point djAre redundant and should be eliminated.
6. The method for predicting the pedestrian motion based on the multi-person posture detection of the improved top-down method according to claim 1, wherein the specific implementation process of the step 6 is as follows: frame the human body regionMapping to the original image coordinate, namely processing the original image coordinate by adopting an inverse space transformation network STDN to obtain a high-quality area frame in the original image coordinate
Wherein
[γ1γ2]=[θ1θ2]-1
γ3=-1×[γ1γ2]θ3
Wherein, theta1,θ2,θ3All reflect the human body region frameVector coefficient of coordinate relation before and after transformation, [ gamma ]1γ2γ3]Is [ theta ] of1θ2θ3]Transposing;
[θ1θ2]the calculation method comprises the following steps:
θ3the calculation method comprises the following steps:
wherein, W represents a matrix formed by each dimension of the input layer and the output layer of the inverse space transformation network, and J (W, b) represents the position of the pedestrian boundary box b in the inverse space transformation network.
7. The method for predicting the pedestrian motion based on the multi-person gesture detection of the improved top-down method according to claim 1, wherein in the step 8:
v=u+d=[ux+dxuy+dy]T
wherein v is the new position of the bone point in the next frame, u represents the position of the bone point, (x, y) represents the coordinates of the bone point, and (u) represents the coordinates of the bone pointx,uy) Coordinates representing the bone point of the next frame, dx、dyRespectively representing the distance between the bone point and the next frame bone point, I (x, y) representing the pedestrian boundary box where the current bone point is located, J (x + d)x,y+dy) A pedestrian boundary box, w, representing the location of the next frame bone pointx、wyThe size of the two constants defining the integration window of the optical flow method determines the time complexity and the effect of the algorithm.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910921085.XA CN110751056B (en) | 2019-09-27 | 2019-09-27 | Pedestrian motion prediction method based on improved top-down method multi-person gesture detection |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910921085.XA CN110751056B (en) | 2019-09-27 | 2019-09-27 | Pedestrian motion prediction method based on improved top-down method multi-person gesture detection |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110751056A true CN110751056A (en) | 2020-02-04 |
CN110751056B CN110751056B (en) | 2023-05-23 |
Family
ID=69277188
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910921085.XA Active CN110751056B (en) | 2019-09-27 | 2019-09-27 | Pedestrian motion prediction method based on improved top-down method multi-person gesture detection |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110751056B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112464740A (en) * | 2020-11-05 | 2021-03-09 | 北京科技大学 | Image processing method and system for top-down gesture recognition process |
CN112560700A (en) * | 2020-12-17 | 2021-03-26 | 北京赢识科技有限公司 | Information association method and device based on motion analysis and electronic equipment |
CN117115861A (en) * | 2023-10-19 | 2023-11-24 | 四川弘和数智集团有限公司 | Glove detection method and device, electronic equipment and storage medium |
CN117456612A (en) * | 2023-12-26 | 2024-01-26 | 西安龙南铭科技有限公司 | Cloud computing-based body posture automatic assessment method and system |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110013840A1 (en) * | 2008-03-14 | 2011-01-20 | Masahiro Iwasaki | Image processing method and image processing apparatus |
US20170316578A1 (en) * | 2016-04-29 | 2017-11-02 | Ecole Polytechnique Federale De Lausanne (Epfl) | Method, System and Device for Direct Prediction of 3D Body Poses from Motion Compensated Sequence |
CN109345504A (en) * | 2018-08-07 | 2019-02-15 | 浙江大学 | A kind of bottom-up more people's Attitude estimation methods constrained using bounding box |
CN109815901A (en) * | 2019-01-24 | 2019-05-28 | 杭州电子科技大学 | A kind of more people's Attitude estimation methods based on YOLOv3 algorithm |
US20190206066A1 (en) * | 2017-12-29 | 2019-07-04 | RetailNext, Inc. | Human Analytics Using Fusion Of Image & Depth Modalities |
CN110135375A (en) * | 2019-05-20 | 2019-08-16 | 中国科学院宁波材料技术与工程研究所 | More people's Attitude estimation methods based on global information integration |
-
2019
- 2019-09-27 CN CN201910921085.XA patent/CN110751056B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110013840A1 (en) * | 2008-03-14 | 2011-01-20 | Masahiro Iwasaki | Image processing method and image processing apparatus |
CN101960490A (en) * | 2008-03-14 | 2011-01-26 | 松下电器产业株式会社 | Image processing method and image processing apparatus |
US20170316578A1 (en) * | 2016-04-29 | 2017-11-02 | Ecole Polytechnique Federale De Lausanne (Epfl) | Method, System and Device for Direct Prediction of 3D Body Poses from Motion Compensated Sequence |
US20190206066A1 (en) * | 2017-12-29 | 2019-07-04 | RetailNext, Inc. | Human Analytics Using Fusion Of Image & Depth Modalities |
CN109345504A (en) * | 2018-08-07 | 2019-02-15 | 浙江大学 | A kind of bottom-up more people's Attitude estimation methods constrained using bounding box |
CN109815901A (en) * | 2019-01-24 | 2019-05-28 | 杭州电子科技大学 | A kind of more people's Attitude estimation methods based on YOLOv3 algorithm |
CN110135375A (en) * | 2019-05-20 | 2019-08-16 | 中国科学院宁波材料技术与工程研究所 | More people's Attitude estimation methods based on global information integration |
Non-Patent Citations (4)
Title |
---|
关韬: "基于深度学习的行人姿态识别与行为预判算法研究" * |
许忠雄;张睿哲;石晓军;岳贵杰;刘弋锋;: "深度学习实时多人姿态估计与跟踪" * |
邓益侬; 罗健欣; 金凤林: "基于深度学习的人体姿态估计方法综述" * |
闫芬婷;王鹏;吕志刚;丁哲;乔梦雨;: "基于视频的实时多人姿态估计方法" * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112464740A (en) * | 2020-11-05 | 2021-03-09 | 北京科技大学 | Image processing method and system for top-down gesture recognition process |
CN112560700A (en) * | 2020-12-17 | 2021-03-26 | 北京赢识科技有限公司 | Information association method and device based on motion analysis and electronic equipment |
CN117115861A (en) * | 2023-10-19 | 2023-11-24 | 四川弘和数智集团有限公司 | Glove detection method and device, electronic equipment and storage medium |
CN117115861B (en) * | 2023-10-19 | 2024-01-26 | 四川弘和数智集团有限公司 | Glove detection method and device, electronic equipment and storage medium |
CN117456612A (en) * | 2023-12-26 | 2024-01-26 | 西安龙南铭科技有限公司 | Cloud computing-based body posture automatic assessment method and system |
CN117456612B (en) * | 2023-12-26 | 2024-03-12 | 西安龙南铭科技有限公司 | Cloud computing-based body posture automatic assessment method and system |
Also Published As
Publication number | Publication date |
---|---|
CN110751056B (en) | 2023-05-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110751056A (en) | Pedestrian motion prediction method based on improved top-down method multi-person posture detection | |
CN109325398B (en) | Human face attribute analysis method based on transfer learning | |
JP2015522200A (en) | Human face feature point positioning method, apparatus, and storage medium | |
CN111814816A (en) | Target detection method, device and storage medium thereof | |
CN109522831B (en) | Real-time vehicle detection method based on micro-convolution neural network | |
CN108615241B (en) | Rapid human body posture estimation method based on optical flow | |
WO2020078140A1 (en) | Optimization system and method for motion model | |
Wang et al. | Research on detection technology of various fruit disease spots based on mask R-CNN | |
CN112016594B (en) | Collaborative training method based on field self-adaption | |
CN111524183A (en) | Target row and column positioning method based on perspective projection transformation | |
CN105631899A (en) | Ultrasonic image motion object tracking method based on gray-scale texture feature | |
CN111914756A (en) | Video data processing method and device | |
CN117315547A (en) | Visual SLAM method for solving large duty ratio of dynamic object | |
CN116109673A (en) | Multi-frame track tracking system and method based on pedestrian gesture estimation | |
CN112069997B (en) | Unmanned aerial vehicle autonomous landing target extraction method and device based on DenseHR-Net | |
CN114898034A (en) | Three-dimensional face generation method and device and three-dimensional face replay method and device | |
Xie et al. | A robust hierarchical lip tracking approach for lipreading and audio visual speech recognition | |
CN110532960B (en) | Target-assisted action recognition method based on graph neural network | |
Koumparoulis et al. | Audio-assisted image inpainting for talking faces | |
CN109614898B (en) | Intelligent judging method for train running direction detection | |
CN108830166B (en) | Real-time bus passenger flow volume statistical method | |
Zhang et al. | Real-Time Detection of Small Targets for Video Surveillance Based on MS-YOLOv5 | |
Sun et al. | Research on Vision-based pedestrian detection and tracking algorithm | |
Li | A new physical posture recognition method based on feature complement-oriented convolutional neural network | |
CN109816685A (en) | A kind of position compensation method tracked in video for rapid moving object |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |