WO2021159554A1 - 基于视觉的目标跟踪方法、系统、设备及存储介质 - Google Patents

基于视觉的目标跟踪方法、系统、设备及存储介质 Download PDF

Info

Publication number
WO2021159554A1
WO2021159554A1 PCT/CN2020/076360 CN2020076360W WO2021159554A1 WO 2021159554 A1 WO2021159554 A1 WO 2021159554A1 CN 2020076360 W CN2020076360 W CN 2020076360W WO 2021159554 A1 WO2021159554 A1 WO 2021159554A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
tracking
human body
tracked
current position
Prior art date
Application number
PCT/CN2020/076360
Other languages
English (en)
French (fr)
Inventor
张明
董健
李帅
Original Assignee
睿魔智能科技（深圳）有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 睿魔智能科技（深圳）有限公司 filed Critical 睿魔智能科技（深圳）有限公司
Publication of WO2021159554A1 publication Critical patent/WO2021159554A1/zh
Priority to US17/886,515 priority Critical patent/US20220383518A1/en

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
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • G06T7/248—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving reference images or patches
    • 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
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161—Detection; Localisation; Normalisation
    • G06V40/167—Detection; Localisation; Normalisation using comparisons between temporally consecutive images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/761—Proximity, similarity or dissimilarity measures
    • 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
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161—Detection; Localisation; Normalisation
    • 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
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person
    • G06T2207/30201—Face
    • 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
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • This application relates to the field of image processing technology, and in particular to a vision-based target tracking method, system, equipment and storage medium.
  • Existing vision-based target tracking algorithms usually use correlation filtering algorithms, that is, to obtain the image template of the target to be tracked according to the current image, and then use correlation filtering to calculate the position with the highest matching degree of the target template in the next frame of image, and this position As the location of the target.
  • Correlation filtering tracking algorithms usually use real-time templates for target matching. Although historical templates can be used for optimization to obtain more robust templates, the algorithm does not work well when the target is abrupt and moving quickly. With regular walking and drastic changes in posture, the tracking effect becomes worse.
  • the embodiments of the present application provide a vision-based target tracking method, system, device, and storage medium, which aim to reduce the possibility of target tracking errors, improve tracking effects, and ensure tracking stability.
  • an embodiment of the present application provides a vision-based target tracking method, which includes: Step A: Reading the current frame image, acquiring and saving the face positions and body positions of all characters in the current frame image; Steps B. Obtain the previous frame image and the previous frame position of the target to be tracked, and use the timing tracking algorithm to obtain the current position of the timing tracking target according to the current frame image, the previous frame image, and the previous frame position of the target to be tracked; Step C: Determine whether there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the time sequence tracking target.
  • Step D Track the current position of the target according to the time sequence and others
  • the human body position of the person judges whether the time series tracking target is in the cross state, if yes, go to step E; Step E, judge whether there is a person matching the human body of the target to be tracked among other people that cross the time series tracking target.
  • the current position of the tracking target is taken as the current position of the target, and step F is executed; step F, keeping track, and step G is executed; step G, the current position of the target is taken as the position of the previous frame of the target to be tracked, and the current frame image is taken as the previous frame Image, return to step A.
  • an embodiment of the present application provides a vision-based target tracking system, which includes a face and human body detection module, a timing tracking module, a face matching module, a human body crossing judgment module, a human body matching module, and an update module.
  • the face and human body detection module is used to read the current frame image, obtain and save the face position and human body position of all the characters in the current frame image;
  • the time sequence tracking module is used to obtain the previous frame image and the previous one of the target to be tracked
  • the face matching module is used to determine the timing of tracking the target Whether there is a person matching the face of the target to be tracked outside the area corresponding to the current position, and the current position of the target is determined according to the judgment result;
  • the position judgment time sequence tracking target is in the cross state;
  • the human body matching module is used to judge whether there is a person matching the human body of the target to be tracked among other characters that cross the time sequence tracking target, and determine the current position of the target according to the judgment result;
  • the update module is used to use the current position of the target as the previous frame position of the
  • an embodiment of the present application provides a computer device, which includes a memory and a processor, the memory stores a computer program that can run on the processor, and the processor implements the foregoing when the computer program is executed. Vision-based target tracking method.
  • an embodiment of the present application also provides a computer-readable storage medium that stores a computer program, the computer program includes program instructions, and the program instructions, when executed, implement the aforementioned vision-based target tracking method .
  • the vision-based target tracking method provided in this application combines time series tracking algorithm, face recognition, and human body recognition for target tracking, and combines the time series tracking algorithm to obtain a more reliable target position, reducing the possibility of severe posture changes of the target person It takes face recognition as the first priority of target switching, and uses human body recognition to effectively avoid tracking the wrong person in the cross phase, reducing tracking errors when the target person intersects with other people irregularly, and improving target tracking Accuracy and high robustness.
  • FIG. 1 is a schematic flowchart of a vision-based target tracking method provided by the first embodiment of this application;
  • FIG. 2 is a schematic diagram of a sub-flow of the vision-based target tracking method provided by the first embodiment of this application;
  • FIG. 3 is a schematic flowchart of a vision-based target tracking method provided by the second embodiment of this application.
  • FIG. 4 is a schematic flowchart of a vision-based target tracking method provided by the third embodiment of this application.
  • FIG. 5 is a schematic diagram of a specific flow of a vision-based target tracking method provided by the third embodiment of this application.
  • Fig. 6 is a schematic block diagram of a vision-based target tracking system provided by an embodiment of the application.
  • FIG. 7 is a schematic block diagram of a vision-based target tracking system provided by another embodiment of the application.
  • Fig. 8 is a structural block diagram of a computer device provided by an embodiment of the application.
  • FIG. 1 is a schematic flowchart of a vision-based target tracking method provided by the first embodiment of the present application. As shown in the figure, the method includes the following steps:
  • Step S110 Read the current frame image, acquire and save the face positions and human body positions of all characters in the current frame image.
  • the position represents the coordinates of each end point of the corresponding area in the image, and the corresponding area is rectangular.
  • the corresponding area can be obtained through the combination of the end point coordinates, and the corresponding area in the image can be obtained by the position.
  • the face position is represented by the end point coordinates of the upper left corner and the end point coordinates of the lower right corner of the corresponding area in the current frame image, and the end point coordinates of the upper left corner and the lower right corner of the corresponding area of the human body in the current frame image
  • the endpoint coordinates of represents the position of the human body.
  • Step S120 Obtain the previous frame image and the previous frame position of the target to be tracked.
  • the time-series tracking algorithm is used to obtain the current position of the time-series tracking target .
  • the target to be tracked refers to the person that needs to be tracked
  • the timing tracking target refers to the tracking target obtained based on the timing tracking algorithm
  • the current position of the timing tracking target refers to the human body position of the tracking target obtained based on the timing tracking algorithm.
  • the timing tracking algorithm can use an existing timing filtering target tracking method, for example, Chinese Patent CN110111358A is a target tracking method based on multi-layer timing filtering, which will not be repeated here.
  • step S120 it specifically includes:
  • step S1101 it is determined whether the target to be tracked has been determined, if so, step S120 is executed, otherwise, step S1102 is executed.
  • Step S1102 according to the tracking requirement, determine the target to be tracked among all people, and extract and save the face template and the human body template of the target to be tracked.
  • the step S1102 is specifically: determining the target to be tracked among all people according to the tracking requirements, obtaining the face image and the human body image of the target to be tracked, and using the face recognition model based on the deep neural network to combine the obtained face
  • the image extracts the face template of the target to be tracked, and the human body image of the target to be tracked is extracted using the human body recognition model based on the deep neural network combined with the obtained human body image.
  • Step S1103 Use the position of the human body of the target to be tracked as the current position of the target, and perform step S170.
  • step S120 is executed to obtain the current position of the time-series tracking target; if the target to be tracked is not determined, the target to be tracked is obtained according to the tracking requirements, and the human body position of the target to be tracked is used as the target The current position, and then step S170 is executed to track the target to be tracked.
  • Step S130 Determine whether there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the time-series tracking target, and if not, perform step S140.
  • the position information obtained by the time series tracking is checked to ensure the accuracy of the tracking.
  • Face recognition is used as the first priority of target switching, where the face matching with the target to be tracked refers to matching with the face template of the target to be tracked.
  • the face image of is extracted using a face recognition model based on a deep neural network.
  • Step S140 Determine whether the time series tracking target is in a cross state according to the current position of the time series tracking target and the human body position of other people, and if so, perform step S150.
  • the cross state refers to a state in which areas corresponding to the positions of the human body overlap.
  • the target to be tracked is not in the area outside the current position of the time series tracking target, and then the tracking target is further tracked according to the time series.
  • the current position and the human body position of other characters determine whether the timing tracking target is in a cross state to determine whether the target to be tracked is in the current position of the timing tracking target, so that the face of the person outside the area corresponding to the current position of the timing tracking target is matched in time
  • the cross state of the sequential tracking target within the current position of the sequential tracking target is used to determine the accuracy of the sequential tracking target, which plays a dual role of guarantee.
  • intersection and union ratio represent the ratio between intersection and union.
  • Step S150 Determine whether there is a person matching the human body of the target to be tracked among other characters that cross the time series tracking target. If not, use the current position of the time series tracking target as the target current position, and execute step S160.
  • the current position of the target represents the accurate position of the human body of the target to be tracked.
  • the human body matching with the target to be tracked refers to matching with the human body template of the target to be tracked.
  • the image is extracted using a human body recognition model based on a deep neural network.
  • Step S160 keep track, and execute step S170. After the timing tracking is known to be correct, the target to be tracked is tracked and recorded according to the obtained current position of the target.
  • Step S170 Use the current position of the target as the position of the previous frame of the target to be tracked, and use the current frame image as the previous frame image, and return to step S110.
  • the vision-based target tracking method in the embodiment of the application performs target tracking by combining the timing tracking algorithm, face recognition, and human body recognition multiple target tracking methods, and the timing tracking algorithm is combined to use the timing information of the target to be tracked to obtain more reliable
  • the target position of the target person can reduce the possibility of severe posture changes of the target person, and a reliable target position can be obtained when the target to be tracked is not in the cross state; face recognition is the first priority of target switching to perform sequential tracking of the target Correction, and by using human body recognition to effectively avoid tracking the wrong person in the cross phase, reduce tracking errors when the target person intersects with other people irregularly, improve the accuracy of target tracking, and have high robustness.
  • Fig. 3 is a schematic flowchart of a vision-based target tracking method provided by a second embodiment of the present application.
  • the vision-based target tracking method of this embodiment includes steps S210-S270.
  • step S210 is similar to step S110 in the foregoing embodiment
  • steps S2101-S2103 are similar to steps S1101-S1103 in the foregoing embodiment
  • step S220 is similar to step S120 in the foregoing embodiment
  • steps S260-S270 are similar to the foregoing embodiment.
  • Steps S160-S170 in are similar and will not be repeated here.
  • the different steps S230-S250 in this embodiment will be described in detail below.
  • step S230 is specifically: step S231, determining whether there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the time-series tracking target. If it does not exist, step S240 is executed.
  • step S230 further includes: step S2311, if there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the time-series tracking target, switch the person to the target to be tracked, and obtain the person’s
  • the human body position is taken as the target current position, and step S270 is executed.
  • the timing tracking target refers to the tracking target obtained based on the timing tracking algorithm
  • the current position of the timing tracking target refers to the human body position of the tracking target obtained based on the timing tracking algorithm
  • the timing tracking algorithm can use the existing timing filtering target
  • the tracking method such as Chinese patent CN110111358A, a target tracking method based on multi-layer timing filtering, will not be repeated here.
  • the current position of the target represents the accurate human position of the target to be tracked.
  • the face matching of the target to be tracked refers to matching with the face template of the target to be tracked.
  • the face template represents the face feature template, which is determined by the face image of the target to be tracked. It is obtained by extracting the face recognition model based on deep neural network.
  • the timing tracking target is not the person to be tracked, and the person matching the face template of the target to be tracked needs to be taken as the person to be tracked Target, obtain the human body position of the person as the current position of the target to correct the tracking.
  • step S240 is specifically: step S241, judging whether the time series tracking target is in the cross state according to the current position of the time series tracking target and the human body position of other people; if the time series tracking target is in the cross state, step S250 is executed.
  • the cross state refers to a state in which areas corresponding to the positions of the human body overlap.
  • step S240 further includes: step S2411, if the time-series tracking target is not in the cross state, the current position of the time-series tracking target is taken as the current position of the target, and step S270 is executed.
  • the target to be tracked is not in an area outside the current position of the time-series tracking target, then further follow the time-series tracking target
  • the current position and the human body position of other characters determine whether the timing tracking target is in a cross state to determine whether the target to be tracked is in the current position of the timing tracking target, so that the face of the person outside the area corresponding to the current position of the timing tracking target is matched in time
  • the cross state of the sequential tracking target within the current position of the sequential tracking target is used to determine the accuracy of the sequential tracking target, which plays a dual role of guarantee.
  • the time series tracking target is the correct person to be tracked, and the tracking is correct.
  • the current position of the timing tracking target is the current position of the target.
  • Other people refer to the area corresponding to the human body position of all non-time-series tracking targets in the current frame image and the intersection ratio of the area corresponding to the current position of the time-series tracking target is greater than the preset intersection ratio threshold.
  • Corresponding characters, intersection and union ratio represent the ratio between intersection and union.
  • step S250 specifically includes:
  • Step S251 It is judged whether there is a person matching the human body of the target to be tracked among other characters that cross the time-series tracking target.
  • Step S252 If there is no person matching the human body of the target to be tracked among other characters that cross the time series tracking target, the current position of the time series tracking target is taken as the target current position, and step S260 is executed.
  • step S250 further includes: step S2511, if there is a person matching the human body of the target to be tracked among other people that cross the time-series tracking target, switch the person to the target to be tracked, and obtain the human body position of the person As the target current position, step S270 is executed.
  • the current position of the target represents the accurate human body position of the target to be tracked.
  • the human body matching with the target to be tracked refers to matching with the human body template of the target to be tracked.
  • the human body template represents the human body feature template.
  • the human body recognition model of the neural network is extracted and obtained.
  • the time-series tracking target is not the person that needs to be tracked, and the human body matching the target to be tracked among other people needs to be the target to be tracked.
  • the position of the human body corresponding to the character is taken as the current position of the target.
  • the vision-based target tracking method in the embodiment of the application performs target tracking by combining the timing tracking algorithm, face recognition, and human body recognition multiple target tracking methods, and the timing tracking algorithm is combined to use the timing information of the target to be tracked to obtain more reliable
  • the target position of the target person can reduce the possibility of severe posture changes of the target person, and a reliable target position can be obtained when the target to be tracked is not in the cross state; face recognition is the first priority of target switching to perform sequential tracking of the target Correction, and through the use of human body recognition to effectively avoid tracking the wrong person in the cross phase, reduce tracking errors when the target person intersects with other people irregularly, improve the accuracy of target tracking, and have high robustness.
  • Fig. 4 is a schematic flowchart of a vision-based target tracking method provided by a third embodiment of the present application.
  • the vision-based target tracking method of this embodiment includes steps S310-S370.
  • step S310 is similar to step S210 in the second embodiment
  • steps S3101-S3103 are similar to steps S2101-S2103 in the second embodiment
  • step S320 is similar to step S220 in the second embodiment
  • step S340 is similar to step S220 in the second embodiment.
  • Step S240 in the embodiment is similar
  • steps S360-S370 are similar to steps S260-S270 in the second embodiment, and will not be repeated here.
  • the different steps S330 and S350 in this embodiment will be described in detail below.
  • step S330 specifically includes:
  • Step S331 Obtain the facial features corresponding to all persons not in the area corresponding to the current position of the time-series tracking target, and calculate the facial features and face features corresponding to all the persons not in the area corresponding to the current position of the time-series tracking target.
  • the timing tracking target refers to the tracking target obtained based on the timing tracking algorithm
  • the current position of the timing tracking target refers to the human body position of the tracking target obtained based on the timing tracking algorithm
  • the face template represents the face feature template.
  • the face image is extracted using a face recognition model based on a deep neural network.
  • Persons who acquire the area outside the current position of the time series tracking target can more quickly and conveniently determine whether the target to be tracked is not in the area corresponding to the current position of the time series tracking target, so as to check the position information obtained by the time series tracking to ensure tracking Accuracy.
  • Step S332 Determine whether the face similarity is less than or equal to a preset face similarity threshold, and if the face similarity is less than or equal to the preset face similarity threshold, step S340 is executed.
  • the threshold can quickly determine whether the target to be tracked is not in the area corresponding to the current position of the time series tracking target, so as to determine whether there is a face matching person with the target to be tracked outside the area corresponding to the current position of the time series tracking target.
  • the target to be tracked is in the area corresponding to the current position of the time series tracking target to determine the accuracy of the tracking area.
  • the step S330 further includes: step S3321, if the face similarity is greater than a preset face similarity threshold, obtain the face feature with the greatest face similarity And switch the person corresponding to the person face as the target to be tracked, obtain the human body position of the person as the current position of the target, and execute step S370.
  • the current position of the target represents the accurate human position of the target to be tracked.
  • the target is tracked in the time-series
  • the face similarity between the face feature corresponding to the person who is not in the area corresponding to the current position of the time-series tracking target and the face template of the target to be tracked is greater than the preset face similarity threshold
  • the target is tracked in the time-series
  • the person corresponding to the face corresponding to the face feature with the greatest face similarity is obtained as the target to be tracked, and the target is updated to be the correct target to be tracked, and the corresponding human body position is the current position of the target.
  • the step S340 includes step S3401, step S341, and step S3411.
  • the step S341 is similar to the step S241 in the second embodiment, and the step S3411 is similar to the second embodiment.
  • the step S2411 in the example is similar and will not be repeated here.
  • the step S3401 is executed before the step S341, which is specifically:
  • Step S3401 obtain the body positions of other characters.
  • the step S3401 is specifically: acquiring all regions corresponding to the human body positions of the non-time-series tracking targets in the current frame image, and calculating the intersection ratio with the regions corresponding to the current positions of the time-series tracking targets, The human body position of the person corresponding to the area corresponding to the human body position whose intersection ratio is greater than the preset intersection ratio threshold is acquired, and taken as the human body position of other characters.
  • the intersection and union ratio represents the ratio between the intersection and the union.
  • the non-time-series tracking target refers to the person in all people except the time-series tracking target.
  • the area corresponding to the human body position of the person who is the non-time-series tracking target is calculated and the current position of the time-series tracking target.
  • the intersection ratio of the corresponding area can preliminarily exclude the area corresponding to the human body position of the person who does not cross the area corresponding to the current position of the time-series tracking target, which facilitates the judgment of the intersection state.
  • the larger the intersection ratio the larger the intersection area.
  • step S350 specifically includes:
  • Step S351 Extract the human body features of other people that cross the time-series tracking target, and calculate the human body similarity between the human body features of the other people and the human body template of the target to be tracked.
  • Step S352 Determine whether the human body similarity is less than or equal to a preset human body similarity threshold.
  • Step S353 If the human body similarity is less than or equal to the preset human body similarity threshold, the human body position of the time-series tracking target is taken as the target current position, and step S360 is executed.
  • the human body template represents a human body feature template, which is obtained by extracting a human body image of a target to be tracked using a human body recognition model based on a deep neural network.
  • the current position of the target represents the accurate position of the human body of the target to be tracked.
  • the other characters that cross the timing tracking target are not the ones to be tracked.
  • the timing tracking target is the person that needs to be tracked, and the tracking is correct.
  • the timing tracking target's The current position is the current position of the target, which improves the reliability of tracking.
  • step S352 the method further includes: if the human body similarity is greater than a preset human body similarity threshold, step S3521 is executed.
  • Step S3521 If the human body similarity is greater than the preset human body similarity threshold, obtain the human body corresponding to the human body feature with the greatest human body similarity, switch the person corresponding to the human body as the target to be tracked, and obtain the human body position of the person as the target For the current position, step S370 is executed.
  • the human body similarity of other characters that cross the time series tracking target is greater than the preset human body similarity threshold, it means that the person corresponding to the current position of the time series tracking target is not the person to be tracked, and needs to be obtained from the human body positions of other characters
  • the person corresponding to the human body corresponding to the human body feature with the greatest human body similarity is used as the target to be tracked, and the corresponding human body position is the current position of the target.
  • the vision-based target tracking method provided by the embodiments of the application combines the timing tracking algorithm to use the timing information of the target to be tracked to obtain a more reliable target position, reduce the possibility of severe posture changes of the target person, and can be used when the target to be tracked is not in the intersection. Obtain a reliable target position in the state; and use human body recognition to effectively avoid tracking the wrong person in the cross phase, reduce tracking errors when the target person crosses other people irregularly, improve the accuracy of target tracking, and robustness high.
  • Fig. 6 is a schematic block diagram of a vision-based target tracking system provided by an embodiment of the application.
  • the target tracking system 10 includes a human face detection module 110, a timing tracking module 120, and a human face.
  • the human face and human body detection module 110 is used to read the current frame image, obtain and save the face positions and human body positions of all people in the current frame image.
  • the position represents the coordinates of each end point of the corresponding area in the image, and the corresponding area is rectangular.
  • the corresponding area can be obtained by combining the coordinates of each end point, and the corresponding area in the image can be obtained by the position.
  • the face position is represented by the end point coordinates of the upper left corner and the end point coordinates of the lower right corner of the corresponding area in the current frame image, and the end point coordinates of the upper left corner and the lower right corner of the corresponding area of the human body in the current frame image
  • the endpoint coordinates of represents the position of the human body.
  • the timing tracking module 120 is used to obtain the previous frame image and the previous frame position of the target to be tracked.
  • a timing tracking algorithm is used to obtain the timing Track the current position of the target.
  • the target to be tracked refers to the person to be tracked
  • the timing tracking target refers to the tracking target obtained based on the timing tracking algorithm
  • the current position of the timing tracking target refers to the human body position of the tracking target obtained based on the timing tracking algorithm.
  • the timing tracking algorithm can use an existing timing filtering target tracking method, for example, Chinese Patent CN110111358A is a target tracking method based on multi-layer timing filtering, which will not be repeated here.
  • the face matching module 130 is used to determine whether there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the time-series tracking target, and determine the current position of the target according to the determination result. Specifically, the judging whether there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the time-series tracking target is calculated by calculating all persons not in the area corresponding to the current position of the time-series tracking target correspond to The face similarity between the face feature and the face template of the target to be tracked is judged in combination with the preset face similarity threshold. According to the judgment result, the current position of the target can be determined.
  • the face matching module 130 obtains the facial features corresponding to all persons not in the area corresponding to the current position of the time series tracking target, and calculates all the persons corresponding to the persons not in the area corresponding to the current position of the time series tracking target
  • the face similarity between the face feature and the face template of the target to be tracked is judged whether the face similarity is less than or equal to a preset face similarity threshold. Wherein, if the face similarity is greater than the preset face similarity threshold, the time sequence tracking target is not the correct target to be tracked, and the face corresponding to the facial feature with the greatest face similarity needs to be obtained, and the face
  • the corresponding person is switched to the target to be tracked, and the position of the person's body is acquired as the current position of the target.
  • the face template represents a face feature template, which is obtained by extracting a face image of a target to be tracked using a face recognition model based on a deep neural network. If the human face similarity is less than or equal to the preset human face similarity threshold, the human body intersection judgment module 140 works.
  • the use of the face matching module 130 to obtain the person in the area outside the current position of the time-series tracking target can more quickly and conveniently determine whether the target to be tracked is no longer in the area corresponding to the current position of the time-series tracking target, so as to track the time series.
  • the location information obtained is checked to ensure the accuracy of tracking.
  • the human body crossing judgment module 140 is used for judging whether the time series tracking target is in the crossing state according to the current position of the time series tracking target and the human body position of other people.
  • the cross state refers to a state in which areas corresponding to the positions of the human body overlap.
  • the human body intersection judgment module 140 obtains the regions corresponding to the human body positions of all non-time-series tracking targets in the current frame image, calculates the intersection ratio with the regions corresponding to the current positions of the time-series tracking targets, and obtains the The human body position corresponding to the human body position corresponding to the area whose intersection ratio is greater than the preset intersection ratio threshold is taken as the human body position of other characters; the current position of the tracking target and the human body position of the other characters are judged according to the timing sequence
  • the timing tracking target is in the cross state.
  • the intersection and union ratio represents the ratio between the intersection and the union.
  • the non-time-series tracking target refers to the person in all people except the time-series tracking target.
  • the corresponding area of the human body position of the person who is not the time-series tracking target is calculated and the current time-series tracking target
  • the intersection ratio of the location corresponding area can preliminarily exclude the human body position of the person who does not cross the current position of the time-series tracking target, which facilitates the judgment of the intersection state.
  • the larger the intersection ratio the larger the intersection area.
  • Other people refer to the human body position corresponding areas of all non-time-series tracking targets in the current frame image and the corresponding areas of the current position of the time-series tracking targets whose intersection ratio is greater than the preset intersection ratio threshold. figure. If the timing tracking target is not in the cross state, the current position of the timing tracking target is taken as the current position of the target.
  • the human body cross module 140 is used to determine the cross state of the time series tracking target within the current position of the time series tracking target, which can provide a double guarantee to improve the tracking target. Accuracy.
  • the human body matching module 150 is used to determine whether there is a person matching the human body of the target to be tracked among other people that cross the time-series tracking target, and determine the current position of the target according to the determination result. Specifically, the judging whether there is a person matching the human body of the target to be tracked among other characters that cross the time series tracking target is to calculate the human body of the other person by extracting the human body characteristics of the other people who cross the time series tracking target. The human body similarity between the feature and the human body template of the target to be tracked is judged in combination with the preset human body similarity threshold. According to the judgment result, the current position of the target can be determined.
  • the time-series tracking target is the person to be tracked, and the current position of the time-series tracking target is taken as the current position of the target.
  • the human body template represents a human body feature template, which is obtained by extracting a human body image of a target to be tracked using a human body recognition model based on a deep neural network. If the human body similarity is greater than the preset human body similarity threshold, the human body corresponding to the human body feature with the greatest human body similarity is obtained, the person corresponding to the human body is switched to the target to be tracked, and the human body position of the person is obtained as the target current position.
  • the human body similarity of other characters that cross the time series tracking target is greater than the preset human body similarity threshold, it means that the person corresponding to the current position of the time series tracking target is not the person to be tracked, and needs to be obtained from the human body positions of other characters
  • the person corresponding to the human body corresponding to the human body feature with the greatest human body similarity is used as the target to be tracked, and the corresponding human body position is the current position of the target.
  • the human body matching module 150 can better track the time sequence tracking target in the cross state, and improve the accuracy of tracking.
  • the update module 160 is configured to use the current position of the target as the previous frame position of the target to be tracked, and the current frame image as the previous frame image.
  • the face and human detection module 110 reads the current frame image, acquires and saves the face positions and human body positions of all the characters in the current frame image; the timing tracking module 120 acquires the previous frame image and the target to be tracked According to the current frame image, the previous frame image, and the previous frame position of the target to be tracked, the time sequence tracking algorithm is used to obtain the current position of the time sequence tracking target; the face matching module 130 obtains all that are not in the time sequence The face features corresponding to the person in the area corresponding to the current position of the tracking target are calculated, and the difference between the face features corresponding to all persons not in the area corresponding to the current position of the tracking target and the face template of the target to be tracked is calculated. Facial similarity.
  • the face similarity is less than or equal to a preset face similarity threshold, so as to judge whether there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the time-series tracking target.
  • a preset face similarity threshold there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the timing tracking target, and the timing tracking target is incorrect to be tracked.
  • the update module 160 works, and the update module 160
  • the current position of the target is taken as the previous frame position of the target to be tracked, and the current frame image is taken as the previous frame image; and the human face detection module 110, the timing tracking module 120, and the face matching module 130 work in sequence; If the face similarity is less than or equal to the preset face similarity threshold, there is no person matching the face of the target to be tracked outside the area corresponding to the current position of the time-series tracking target, and the human body crossing judgment module 140 works; the human body crossing judgment The module 140 judges whether the time series tracking target is in the cross state according to the current position of the time series tracking target and the human body position of other characters.
  • the current position of the time series tracking target is used as the target current position, and the module 160 is updated.
  • Work, and the face and human detection module 110, the timing tracking module 120, the face matching module 130, and the human cross judging module 140 work in sequence; if the timing tracking target is in the cross state, the human body matching module 150 works; the human body matching module 150 extracts and Time sequence tracking the human body characteristics of other people where the target crosses, calculate the human body similarity between the human body characteristics of the other people and the human body template of the target to be tracked, and determine the current position of the target according to the human body similarity to determine the time sequence Whether there is a person matching the human body of the target to be tracked among other people whose tracking target crosses, if the human body similarity is greater than the preset human body similarity threshold, there are other people who cross the time-series tracking target with the target to be tracked.
  • the vision-based target tracking system of the embodiment of the present application combines the timing tracking algorithm, face recognition, and human body recognition to perform target tracking, and combines the timing tracking algorithm to obtain a more reliable target position, reducing the possibility of severe posture changes of the target person , Regard face recognition as the first priority of target switching, and avoid tracking the wrong person in the cross phase by using human body recognition, reduce the tracking error when the target person intersects with other people irregularly, and improve the accuracy of target tracking , And high robustness.
  • Fig. 7 is a schematic block diagram of a vision-based target tracking system provided by another embodiment of the present application.
  • the target tracking system 20 adds an initialization module 270 based on the above embodiment.
  • the initialization module 270 is used to determine the target to be tracked among all people according to the tracking requirements, and extract and save the target to be tracked.
  • the face template and the human body template of the target use the human body position of the target to be tracked as the current position of the target.
  • the initialization module 270 determines the target to be tracked among all people, obtains the face image and the human body image of the target to be tracked, and extracts the target to be tracked by combining the obtained face image with the face recognition model based on the deep neural network.
  • Use the human body image based on the deep neural network to extract the human body template of the target to be tracked, and use the human body position of the target to be tracked as the current position of the target.
  • the face and human body detection module 210 reads the current frame image, acquires and saves the face positions and human body positions of all people in the current frame image; the initialization module 270 determines the target to be tracked in all people according to requirements , Obtain the face image and human body image of the target to be tracked, and extract the face template of the target to be tracked by combining the face recognition model based on deep neural network with the obtained face image, and use the combination of the human body recognition model based on deep neural network to obtain The human body image extracts the human body template of the target to be tracked, and uses the human body position of the target to be tracked as the current position of the target; the update module 260 uses the current position of the target as the previous frame position of the target to be tracked and the current frame image as the previous frame image; The face and human body detection module reads the current frame image again, acquires and saves the face positions and human body positions of all people in the current frame image; the timing tracking module 220 acquires the previous frame image and the previous frame position
  • the face similarity is less than or equal to a preset face similarity threshold, so as to judge whether there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the time-series tracking target.
  • a preset face similarity threshold there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the time-series tracking target, and the time-series tracking target is not a person to be tracked.
  • the face corresponding to the facial feature with the greatest face similarity switch the person corresponding to the face as the target to be tracked, obtain the human body position of the person as the current position of the target, the update module 260 works, and the human face
  • the detection module 210, the timing tracking module 220, and the face matching module 250 work in sequence; if the face similarity is less than or equal to the preset face similarity threshold, the human body crossing judgment module 240 works; the human body crossing judgment module 240 works according to The current position of the time series tracking target and the human body position of other people determine whether the time series tracking target is in the cross state.
  • the current position of the time series tracking target is taken as the target current position, and the update module 260 works, and
  • the face and human body detection module 210, the timing tracking module 220, the face matching module 250, and the human body intersection judgment module 240 work in sequence; if the timing tracking target is in the cross state, the human body matching module 250 works; the human body matching module 250 extracts and timing tracking targets
  • the human body characteristics of other people that cross calculate the human body similarity between the human body characteristics of the other people and the human body template of the target to be tracked, and determine the current position of the target according to the human body similarity to determine and track the target occurrence in time sequence Whether there is a person matching the human body of the target to be tracked among the other crossed characters.
  • the human body similarity is greater than the preset human body similarity threshold, there is a person matching the human body of the target to be tracked among other characters that cross the time series tracking target, and the time series tracking target is not the person that needs to be tracked, and the human body similarity is maximized.
  • a timing tracking module, a face matching module, a human body crossing judgment module, and a human body matching module are set to combine the timing tracking algorithm, face recognition, and human body recognition multiple target tracking methods for target tracking, combined with timing tracking
  • the algorithm uses the timing information of the target to be tracked to obtain a more reliable target position, reduces the possibility of severe posture changes of the target person, and can obtain a reliable target position when the target to be tracked is not in a cross state; face recognition is used as the target switch
  • the first priority is to correct the timing tracking target, and use human body recognition to effectively avoid tracking the wrong person in the cross phase, reduce tracking errors when the target person intersects with other people irregularly, and improve the accuracy of target tracking It has high performance and high robustness.
  • the above vision-based target tracking system can be implemented in the form of a computer program, and the computer program can be run on a computer device as shown in FIG. 8.
  • the computer device 30 may be a terminal. It can also be a server, where the terminal can be an electronic device with communication functions such as a tablet computer, a notebook computer, and a desktop computer.
  • the server can be an independent server or a server cluster composed of multiple servers.
  • the computer device 30 includes a processor 302, a memory, and a network interface 305 connected through a system bus 301, where the memory may include a non-volatile storage medium 303 and an internal memory 304.
  • the non-volatile storage medium 303 can store an operating system 3031 and a computer program 3032.
  • the computer program 3032 includes program instructions. When the program instructions are executed, the processor 302 can execute a vision-based target tracking method.
  • the processor 302 is used to provide calculation and control capabilities to support the operation of the entire computer device 30.
  • the internal memory 304 provides an environment for the operation of the computer program 3032 in the non-volatile storage medium 303.
  • the processor 302 can execute a vision-based target tracking method.
  • the network interface 305 is used for network communication with other devices.
  • the specific computer device 30 may include more or fewer parts than shown in the figure, or combine certain parts, or have a different arrangement of parts.
  • the processor 302 is configured to run a computer program 3032 stored in a memory to implement a vision-based target tracking method.
  • the target tracking method includes: step A, reading the current frame image, acquiring and saving the current frame The face positions and human body positions of all characters in the image; step B, obtaining the previous frame image and the previous frame position of the target to be tracked, according to the current frame image, the previous frame image and the previous frame position of the target to be tracked , Use the time series tracking algorithm to obtain the current position of the time series tracking target; step C, determine whether there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the time series tracking target, if not, perform step D; step D.
  • step E determines whether the time series tracking target is in the cross state, if yes, perform step E; If the person matching the human body of the tracking target does not exist, use the current position of the timing tracking target as the target current position, and perform step F; step F, keep tracking, perform step G; step G, use the current position of the target as the target to be tracked Take the current frame image as the previous frame image, and return to step A.
  • the computer program stored therein is not limited to the above method operations, and can also perform related operations in the vision-based target tracking method provided by any embodiment of the present application.
  • the processor 302 may be a central processing unit (Central Processing Unit, CPU), and the processor 302 may also be other general-purpose processors, digital signal processors (Digital Signal Processors, DSPs), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor.
  • the computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium.
  • the program instructions are executed by at least one processor in the computer system to implement the process steps of the foregoing method embodiments.
  • the storage medium may be a computer-readable storage medium.
  • the storage medium stores a computer program, where the computer program includes program instructions.
  • the processor realizes a vision-based target tracking method.
  • the target tracking method includes: step A, reading the current frame image, acquiring and saving the face positions of all characters in the current frame image and Human body position; step B. Obtain the previous frame image and the previous frame position of the target to be tracked.
  • the timing tracking algorithm is used to obtain the timing tracking target Step C. Determine whether there is a person matching the face of the target to be tracked outside the area corresponding to the current position of the timing tracking target.
  • Step D tracking the target according to the timing
  • the current position and the human body position of other people determine whether the time series tracking target is in the cross state, if yes, perform step E; step E, determine whether there is a person matching the human body of the target to be tracked among other people that cross the time series tracking target, if If it does not exist, use the current position of the time-series tracking target as the current position of the target, and perform step F; step F, keep tracking, and perform step G; step G, use the current position of the target as the previous frame position of the target to be tracked, and set the current frame The image is taken as the previous frame image, and step A is returned to.
  • the program instructions included in the computer program stored therein are not limited to the above method operations, and can also perform related operations in the vision-based target tracking method provided by any embodiment of the present application .
  • the storage medium may be a portable hard disk, a read-only memory (Read-Only Memory, ROM), a magnetic disk, or an optical disk, and other computer-readable storage media that can store program codes.
  • ROM Read-Only Memory
  • the storage medium may be a portable hard disk, a read-only memory (Read-Only Memory, ROM), a magnetic disk, or an optical disk, and other computer-readable storage media that can store program codes.
  • the units in the system of the embodiment of the present application may be combined, divided, and deleted according to actual needs.
  • the functional units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Human Computer Interaction (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

一种基于视觉的目标跟踪方法、系统、设备及存储介质，该方法包括：读取当前帧图像，获取并保存当前帧图像中的所有人物的人脸位置及人体位置（S110）；获取前一帧图像及待跟踪目标的前一帧位置，根据所述当前帧图像、前一帧图像及待跟踪目标的前一帧位置，采用时序跟踪算法获取时序跟踪目标的当前位置（S120）；判断在在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物，若不存在，执行步骤S140（S130）；根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序目标是否处于交叉状态，若是，执行步骤S150（S140）；判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，若不存在，则将时序跟踪目标的当前位置作为目标当前位置，执行步骤S160（S150）；保持跟踪（S160）；将目标当前位置作为待跟踪目标的前一帧位置，将当前帧图像作为前一帧图像（S170），返回执行上述步骤。该方法能降低目标跟踪错误的可能性，确保跟踪的稳定性。

Description

基于视觉的目标跟踪方法、系统、设备及存储介质
本申请是以申请号为202010095228.9、申请日为2020年2月14日的中国专利申请为基础，并主张其优先权，该申请的全部内容在此作为整体引入本申请中。
技术领域
本申请涉及图像处理技术领域，尤其涉及一种基于视觉的目标跟踪方法、系统、设备及存储介质。
背景技术
现有的基于视觉的目标跟踪算法通常采用相关滤波算法，即根据当前图像获取需要跟踪目标的图像模板，然后采用相关滤波计算得到下一帧图像中和目标模板匹配度最高的位置，将该位置作为目标所在的位置。
相关滤波跟踪算法通常采用实时模板进行目标匹配，虽然可采用历史模板进行优化得到更具有鲁棒性的模板，但是该算法在目标突变、快速移动的情况下效果不好，则使得当人体出现无规律的走动再加上姿态的剧烈变化的情况下，跟踪效果变差。
申请内容
本申请实施例提供了一种基于视觉的目标跟踪方法、系统、设备及存储介质，旨在降低目标跟踪错误的可能性，提高跟踪效果，确保跟踪的稳定性。
第一方面，本申请实施例提供了一种基于视觉的目标跟踪方法，其包括：步骤A、读取当前帧图像，获取并保存当前帧图像中的所有人物的人脸位置及人体位置；步骤B、获取前一帧图像及待跟踪目标的前一帧位置，根据所述当前帧图像、前一帧图像及待跟踪目标的前一帧位置，采用时序跟踪算法获取时序跟踪目标的当前位置；步骤C、判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物，若不存在，执行步骤D；步骤D、根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态，若是，执行步骤E；步骤E、判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，若不存在，将时序跟踪目标的当前 位置作为目标当前位置，执行步骤F；步骤F、保持跟踪，执行步骤G；步骤G、将目标当前位置作为待跟踪目标的前一帧位置，将当前帧图像作为前一帧图像，返回执行步骤A。
第二方面，本申请实施例提供了一种基于视觉的目标跟踪系统，其包括人脸人体检测模块、时序跟踪模块、人脸匹配模块、人体交叉判断模块、人体匹配模块及更新模块，所述人脸人体检测模块用于读取当前帧图像，获取并保存当前帧图像中的所有人物的人脸位置及人体位置；所述时序跟踪模块用于获取前一帧图像及待跟踪目标的前一帧位置，根据所述当前帧图像、前一帧图像及待跟踪目标的前一帧位置，采用时序跟踪算法获取时序跟踪目标的当前位置；所述人脸匹配模块用于判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物，并根据判断结果确定目标当前位置；所述人体交叉判断模块用于根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态；所述人体匹配模块用于判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，并根据判断结果确定目标当前位置；所述更新模块用于将目标当前位置作为待跟踪目标的前一帧位置，将当前帧图像作为前一帧图像。
第三方面，本申请实施例提供了一种计算机设备，其包括存储器和处理器，所述存储器上存储有可在处理器运行的计算机程序，所述处理器执行所述计算机程序时实现上述的基于视觉的目标跟踪方法。
第四方面，本申请实施例还提供了一种计算机可读存储介质，其存储有计算机程序，所述计算机程序包括程序指令，所述程序指令当被执行时实现上述的基于视觉的目标跟踪方法。
本申请提供的一种基于视觉的目标跟踪方法将时序跟踪算法、人脸识别及人体识别结合在一起进行目标跟踪，结合时序跟踪算法获得更可靠的目标位置，降低目标人物发生剧烈姿态变化的可能性，将人脸识别作为目标切换的第一优先级，并通过利用人体识别以有效避免交叉阶段跟踪到错误的人物，降低当目标人物与其他人物无规律交叉时的跟踪错误，提高目标跟踪的准确性，且鲁棒性高。
附图说明
为了更清楚地说明本申请实施例技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍。
图1为本申请第一实施例提供的基于视觉的目标跟踪方法的流程示意图；
图2为本申请第一实施例提供的基于视觉的目标跟踪方法的子流程示意图；
图3为本申请第二实施例提供的基于视觉的目标跟踪方法的流程示意图；
图4为本申请第三实施例提供的基于视觉的目标跟踪方法的流程示意图；
图5为本申请第三实施例提供的基于视觉的目标跟踪方法的具体流程示意图；
图6为本申请实施例提供的基于视觉的目标跟踪系统的示意性框图；
图7为本申请另一实施例提供的基于视觉的目标跟踪系统的示意性框图；
图8为本申请实施例提供的计算机设备的结构框图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。
应当理解，当在本说明书和所附权利要求书中使用时，术语“包括”和“包含”指示所描述特征、整体、步骤、操作、元素和/或组件的存在，但并不排除一个或多个其他特征、整体、步骤、操作、元素、组件和/或其集合的存在或添加。在此本申请说明书中使用的术语仅仅是出于描述特定实施例的目的而并不意在限制本申请。如在本申请说明书和所附权利要求书中所使用的那样，除非上下文清楚地指明其他情况，否则单数形式的“一”、“一个”及“该”意在包括复数形式。在本申请说明书和所附权利要求书中使用的属于“和/或”是指相关联列出的项中的一个或多个的任何组合及所有可能组合，并且包括这些组合。
请参阅图1，图1是本申请第一实施例提供的一种基于视觉的目标跟踪方法的流程示意图。如图所示，该方法包括以下步骤：
步骤S110、读取当前帧图像，获取并保存当前帧图像的所有人物的人脸位置及人体位置。其中，位置代表对应区域于图像中的各端点坐标，对应区域呈矩形，通过各端点坐标组合可获得对应的区域，则可通过位置获取于图像中的对应 区域。具体地，以人脸于当前帧图像中的对应区域的左上角的端点坐标及右下角的端点坐标表示人脸位置，以人体于当前帧图像中的对应区域的左上角的端点坐标及右下角的端点坐标表示人体位置。
步骤S120、获取前一帧图像及待跟踪目标的前一帧位置，根据所述当前帧图像、前一帧图像及待跟踪目标的前一帧位置，采用时序跟踪算法获取时序跟踪目标的当前位置。其中，待跟踪目标是指需要跟踪的人物，时序跟踪目标是指基于时序跟踪算法获得的跟踪目标，时序跟踪目标的当前位置是指基于时序跟踪算法获得的跟踪目标的人体位置。利用时序跟踪算法以跟踪所述待跟踪目标的时序信息可预测获取更可靠的位置信息。其中，所述时序跟踪算法可采用现有的时序滤波的目标跟踪方法，例如中国专利CN110111358A一种基于多层时序滤波的目标跟踪方法，此不赘述。
具体地，在一些实施例中，如图2所示，所述步骤S120前还具体包括：
步骤S1101、判断是否已确定待跟踪目标，若是，执行步骤S120，否则，执行步骤S1102。
步骤S1102、根据跟踪需求，在所有人物中确定待跟踪目标，提取并保存待跟踪目标的人脸模板及人体模板。
优选地，所述步骤S1102具体为：根据跟踪需求，在所有人物中确定待跟踪目标，获取待跟踪目标的人脸图像和人体图像，利用基于深度神经网络的人脸识别模型结合获得的人脸图像提取待跟踪目标的人脸模板，利用基于深度神经网络的人体识别模型结合获得的人体图像提取待跟踪目标的人体模板。
步骤S1103、将待跟踪目标的人体位置作为目标当前位置，执行步骤S170。
本实施例中，若已确定待跟踪目标，则执行步骤S120以获取时序跟踪目标的当前位置；若未确定待跟踪目标，则根据跟踪需求获取待跟踪目标，将待跟踪目标的人体位置作为目标当前位置，然后执行步骤S170，以对待跟踪目标进行跟踪。
步骤S130、判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物，若不存在，执行步骤S140。通过判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物以判断是否跟 踪错误，从而对时序跟踪获得的位置信息进行检验，确保跟踪的准确性，同时，利用人脸识别作为目标切换的第一优先级，其中，与待跟踪目标的人脸匹配是指与待跟踪目标的人脸模板匹配，所述人脸模板表示人脸特征模板，由待跟踪目标的人脸图像利用基于深度神经网络的人脸识别模型提取获得。
步骤S140、根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态，若是，执行步骤S150。其中，交叉状态是指人物的人体位置对应的区域之间发生重叠的状态。
若在时序跟踪目标的当前位置对应的区域外不存在与待跟踪目标的人脸匹配的人物，则待跟踪目标不在时序跟踪目标的当前位置外的区域，则进一步地根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态，以确定待跟踪目标是否处于时序跟踪目标的当前位置，这样通过时序跟踪目标的当前位置对应的区域外的人物的人脸匹配及时序跟踪目标的当前位置内的时序跟踪目标的交叉状态来判断时序跟踪目标的准确性，起到双重保障的作用。其他人物是指当前帧图像中所有的非时序跟踪目标的人体位置对应的区域与所述时序跟踪目标的当前位置对应的区域的交并比大于预设的交并比阈值的人体位置对应区域所对应的人物，交并比代表交集与并集之间的比值。
步骤S150、判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，若不存在，将时序跟踪目标的当前位置作为目标当前位置，执行步骤S160。
其中，所述目标当前位置代表待跟踪目标的准确的人体位置，与待跟踪目标的人体匹配是指与待跟踪目标的人体模板匹配，所述人体模板表示人体特征模板，由待跟踪目标的人体图像利用基于深度神经网络的人体识别模型提取获得。通过判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，以判断其他人物是否为待跟踪目标，提高跟踪目标的准确性。当与时序跟踪目标发生交叉的其他人物中不存在与待跟踪目标的人体匹配的人物，则与时序跟踪目标发生交叉的其他人物并非需要跟踪的人物，时序跟踪目标即为需要跟踪的人物，则跟踪无误，此时，时序跟踪目标的当前位置即为目标 当前位置，提高跟踪的可靠性。
步骤S160、保持跟踪，执行步骤S170。在已知时序跟踪无误后，根据获得的目标当前位置以对待跟踪目标进行跟踪记录。
步骤S170、将目标当前位置作为待跟踪目标的前一帧位置，将当前帧图像作为前一帧图像，返回执行步骤S110。
本申请实施例的基于视觉的目标跟踪方法通过将时序跟踪算法、人脸识别及人体识别多种目标跟踪方式结合在一起进行目标跟踪，结合时序跟踪算法以利用待跟踪目标的时序信息获得更可靠的目标位置，降低目标人物发生剧烈姿态变化的可能性，可在待跟踪目标不处于交叉状态时获取可靠的目标位置；将人脸识别作为目标切换的第一优先级，以对时序跟踪目标进行修正，并通过利用人体识别以有效避免交叉阶段跟踪到错误的人物，降低当目标人物与其他人物无规律交叉时的跟踪错误，提高目标跟踪的准确性，且鲁棒性高。
图3是本申请第二实施例提供的基于视觉的目标跟踪方法的流程示意图。如图3所示，本实施例的基于视觉的目标跟踪方法包括步骤S210-S270。其中，步骤S210与上述实施例中的步骤S110类似，步骤S2101-S2103与上述实施例中的步骤S1101-S1103类似，步骤S220与上述实施例中的步骤S120类似，步骤S260-S270与上述实施例中的步骤S160-S170类似，在此不再赘述。下面详细说明本实施例中不同的步骤S230-S250。
在本实施例中，所述步骤S230具体为：步骤S231、判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物。若不存在，执行步骤S240。
具体地，所述步骤S230还包括：步骤S2311、若在时序跟踪目标的当前位置对应的区域外存在与待跟踪目标的人脸匹配的人物，将该人物切换为待跟踪目标，获取该人物的人体位置作为目标当前位置，执行步骤S270。
其中，时序跟踪目标是指基于时序跟踪算法获得的跟踪目标，时序跟踪目标的当前位置是指基于时序跟踪算法获得的跟踪目标的人体位置，所述时序跟踪算法可采用现有的时序滤波的目标跟踪方法，例如中国专利CN110111358A一种基于多层时序滤波的目标跟踪方法，此不赘述。目标当前位置代表待跟踪目标的 准确的人体位置，与待跟踪目标的人脸匹配是指与待跟踪目标的人脸模板匹配，人脸模板表示人脸特征模板，由待跟踪目标的人脸图像利用基于深度神经网络的人脸识别模型提取获得。当在时序跟踪目标的当前位置对应的区域外存在与待跟踪目标的人脸匹配的人物，则时序跟踪目标并非需要跟踪的人物，需将与待跟踪目标的人脸模板匹配的人物作为待跟踪目标，获取该人物的人体位置作为目标当前位置，以更正跟踪。
在本实施例中，步骤S240具体为：步骤S241、根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态，若时序跟踪目标处于交叉状态，执行步骤S250。其中，交叉状态是指人物的人体位置对应的区域之间发生重叠的状态。
具体地，所述步骤S240还包括：步骤S2411、若时序跟踪目标不处于交叉状态，则将时序跟踪目标的当前位置作为目标当前位置，执行步骤S270。
若在时序跟踪目标的当前位置对应的区域外不存在与待跟踪目标的人脸匹配的人物，即待跟踪目标不在时序跟踪目标的当前位置外的区域，则进一步地根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态，以确定待跟踪目标是否处于时序跟踪目标的当前位置，这样通过时序跟踪目标的当前位置对应的区域外的人物的人脸匹配及时序跟踪目标的当前位置内的时序跟踪目标的交叉状态来判断时序跟踪目标的准确性，起到双重保障的作用。若在时序跟踪目标的当前位置对应的区域外不存在与待跟踪目标的人脸匹配的人物，且时序跟踪目标又不处于交叉状态，则时序跟踪目标为正确的需要跟踪的人物，跟踪无误，此时，时序跟踪目标的当前位置即为目标当前位置。其他人物是指当前帧图像中所有的非时序跟踪目标的人体位置对应的区域与所述时序跟踪目标的当前位置对应的区域的交并比大于预设的交并比阈值的人体位置对应区域所对应的人物，交并比代表交集与并集之间的比值。
在本实施例中，步骤S250具体包括：
步骤S251、判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物。
步骤S252、若与时序跟踪目标发生交叉的其他人物中不存在与待跟踪目标的人 体匹配的人物，则将时序跟踪目标的当前位置作为目标当前位置，执行步骤S260。
具体地，所述步骤S250还包括：步骤S2511、若与时序跟踪目标发生交叉的其他人物中存在与待跟踪目标的人体匹配的人物，将该人物切换为待跟踪目标，获取该人物的人体位置作为目标当前位置，执行步骤S270。
其中，目标当前位置代表待跟踪目标的准确的人体位置，与待跟踪目标的人体匹配是指与待跟踪目标的人体模板匹配，人体模板表示人体特征模板，由待跟踪目标的人体图像利用基于深度神经网络的人体识别模型提取获得。通过判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，提高跟踪目标的准确性。当与时序跟踪目标发生交叉的其他人物中不存在与待跟踪目标的人体匹配的人物，则时序跟踪目标即为需要跟踪的人物，则跟踪无误，此时，时序跟踪目标的当前位置即为目标当前位置，提高跟踪的可靠性。当与时序跟踪目标发生交叉的其他人物中存在与待跟踪目标的人体匹配的人物，则时序跟踪目标并非需要跟踪的人物，需将其他人物中与待跟踪目标的人体匹配人物作为待跟踪目标，该人物对应的人体位置作为目标当前位置。
本申请实施例的基于视觉的目标跟踪方法通过将时序跟踪算法、人脸识别及人体识别多种目标跟踪方式结合在一起进行目标跟踪，结合时序跟踪算法以利用待跟踪目标的时序信息获得更可靠的目标位置，降低目标人物发生剧烈姿态变化的可能性，可在待跟踪目标不处于交叉状态时获取可靠的目标位置；将人脸识别作为目标切换的第一优先级，以对时序跟踪目标进行修正，并通过利用人体识别以有效避免交叉阶段跟踪到错误的人物，降低当目标人物与其他人物无规律交叉时的跟踪错误，提高目标跟踪的准确性，且鲁棒性高。
图4是本申请第三实施例提供的基于视觉的目标跟踪方法的流程示意图。如图4所示，本实施例的基于视觉的目标跟踪方法包括步骤S310-S370。其中，步骤S310与第二实施例中的步骤S210类似，步骤S3101-S3103与第二实施例中的步骤S2101-S2103类似，步骤S320与第二实施例中的步骤S220类似，步骤S340与第二实施例中的步骤S240类似，步骤S360-S370与第二实施例中的步骤S260-S270类似，在此不再赘述。下面详细说明本实施例中不同的步骤步骤S330及步骤S350。
在本实施例中，所述步骤S330具体包括：
步骤S331、获取所有不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征，计算所有不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征与待跟踪目标的人脸模板之间的人脸相似度。其中，时序跟踪目标是指基于时序跟踪算法获得的跟踪目标，时序跟踪目标的当前位置是指基于时序跟踪算法获得的跟踪目标的人体位置，人脸模板表示人脸特征模板，由待跟踪目标的人脸图像利用基于深度神经网络的人脸识别模型提取获得。获取时序跟踪目标的当前位置外的区域的人物可以更为快速和方便地判断出待跟踪目标是否不在所述时序跟踪目标的当前位置对应区域，从而对时序跟踪获得的位置信息进行检验，确保跟踪的准确性。
步骤S332、判断所述人脸相似度是否小于等于预设的人脸相似阈值，若所述人脸相似度小于等于预设的人脸相似阈值，执行步骤S340。
通过获取不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征，并判断其与待跟踪目标的人脸模板之间的人脸相似度是否小于等于预设的人脸相似阈值，可快速判断待跟踪目标是否不在所述时序跟踪目标的当前位置对应的区域内，从而判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配人物。若所有不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征与待跟踪目标的人脸模板之间的人脸相似度均小于等于预设的人脸相似阈值，则在时序跟踪目标的当前位置对应的区域外不存在与待跟踪目标的人脸匹配的人物，则待跟踪目标在所述时序跟踪目标的当前位置对应的区域内，确定跟踪区域的准确。
在一些实施例，如图4和图5所示，所述步骤S330还包括：步骤S3321，若所述人脸相似度大于预设的人脸相似阈值，获取人脸相似度最大的人脸特征对应的人脸，并将该人脸对应的人物切换为待跟踪目标，获取该人物的人体位置作为目标当前位置，执行步骤S370。目标当前位置代表待跟踪目标的准确的人体位置。当不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征与待跟踪目标的人脸模板之间的人脸相似度大于预设的人脸相似阈值，则在时序跟踪目标的当前位置对应的区域外存在与待跟踪目标的人脸匹配的人物，即 代表所述时序跟踪目标的当前位置对应的人物并非需要跟踪的人物，需从人脸相似度大于人脸相似阈值的对应的人脸中获取人脸相似度最大的人脸特征对应的人脸所对应的人物作为待跟踪目标，以更新为正确的待跟踪目标，且其对应的人体位置为目标当前位置。
具体地，在一些实施例中，如图5所示，所述步骤S340包括步骤S3401、步骤S341及步骤S3411，所述步骤S341与第二实施例中的步骤S241类似，步骤S3411与第二实施例中的步骤S2411类似，此不赘述，所述步骤S3401于步骤S341前执行，其具体为：
步骤S3401、获取其他人物的人体位置。
在一些实施例中，所述步骤S3401具体为：获取当前帧图像中所有的非时序跟踪目标的人体位置对应的区域，计算其与所述时序跟踪目标的当前位置对应的区域的交并比，获取所述交并比大于预设的交并比阈值的人体位置对应区域所对应的人物的人体位置，并将其作为其他人物的人体位置。
交并比代表交集与并集之间的比值，非时序跟踪目标是指所有人物中除时序跟踪目标外的人物，计算非时序跟踪目标的人物的人体位置对应的区域与时序跟踪目标的当前位置对应区域的交并比可初步排除与时序跟踪目标的当前位置对应的区域不交叉的人物的人体位置对应的区域，便于交叉状态的判断，交并比越大，交叉面积越大。其他人物是指当前帧图像中所有的非时序跟踪目标的人体位置对应的区域与所述时序跟踪目标的当前位置对应的区域的交并比大于预设的交并比阈值的人体位置对应区域所对应的人物。
在本实施例中，所述步骤S350具体包括：
步骤S351、提取与时序跟踪目标发生交叉的其他人物的人体特征，计算所述其他人物的人体特征与待跟踪目标的人体模板之间的人体相似度。
步骤S352、判断所述人体相似度是否小于等于预设的人体相似阈值。
步骤S353、若所述人体相似度小于等于预设的人体相似阈值，将时序跟踪目标的人体位置作为目标当前位置，执行步骤S360。其中，所述人体模板表示人体特征模板，由待跟踪目标的人体图像利用基于深度神经网络的人体识别模型提取获得。所述目标当前位置代表待跟踪目标准确的人体位置。通过判断所述其 他人物的人体特征与待跟踪目标的人体模板之间的人体相似度以判断其他人物是否为待跟踪目标，提高跟踪目标的准确性。当人体相似度小于等于预设的人体相似阈值，则与时序跟踪目标发生交叉的其他人物并非需要跟踪的人物，时序跟踪目标即为需要跟踪的人物，则跟踪无误，此时，时序跟踪目标的当前位置即为目标当前位置，提高跟踪的可靠性。
在一些实施例中，如图4和图5所示，所述步骤S352后还包括：若所述人体相似度大于预设的人体相似阈值，执行步骤S3521。
步骤S3521、若所述人体相似度大于预设的人体相似阈值，获取人体相似度最大的人体特征对应的人体，并将该人体对应的人物切换为待跟踪目标，获取该人物的人体位置作为目标当前位置，执行步骤S370。
当与时序跟踪目标发生交叉的其他人物的人体相似度大于预设的人体相似阈值，则代表所述时序跟踪目标的当前位置对应的人物并非需要跟踪的人物，需从其他人物的人体位置中获取人体相似度最大的人体特征对应的人体所对应的人物作为待跟踪目标，且其对应的人体位置为目标当前位置。
本申请实施例提供的基于视觉的目标跟踪方法结合时序跟踪算法以利用待跟踪目标的时序信息获得更可靠的目标位置，降低目标人物发生剧烈姿态变化的可能性，可在待跟踪目标不处于交叉状态时获取可靠的目标位置；并通过利用人体识别以有效避免交叉阶段跟踪到错误的人物，降低当目标人物于其他人物无规律交叉时的跟踪错误，提高目标跟踪的准确性，且鲁棒性高。
图6为本申请实施例提供的一种基于视觉的目标跟踪系统的示意性框图。如图6所示，对应于以上基于视觉的目标跟踪方法，本申请提供一种基于视觉的目标跟踪系统10，所述目标跟踪系统10包括人脸人体检测模块110、时序跟踪模块120、人脸匹配模块130、人体交叉判断模块140、人体匹配模块150及更新模块160。所述人脸人体检测模块110用于读取当前帧图像，获取并保存当前帧图像中的所有人物的人脸位置及人体位置。其中，位置代表对应区域于图像中的各端点坐标，对应区域呈矩形，通过各端点坐标组合可获得对应的区域，则可通过位置获取于图像中的对应区域。具体地，以人脸于当前帧图像中的对应区域的左上角的端点坐标及右下角的端点坐标表示人脸位置，以人体于当前帧图像中的 对应区域的左上角的端点坐标及右下角的端点坐标表示人体位置。
所述时序跟踪模块120用于获取前一帧图像及待跟踪目标的前一帧位置，根据所述当前帧图像、前一帧图像及待跟踪目标的前一帧位置，采用时序跟踪算法获取时序跟踪目标的当前位置。其中，待跟踪目标是指需要跟踪的人物，时序跟踪目标是指基于时序跟踪算法获得的跟踪目标，时序跟踪目标的当前位置是指基于时序跟踪算法获得的跟踪目标的人体位置。利用时序跟踪算法以跟踪所述待跟踪目标的时序信息可预测获取更可靠的位置信息。其中，所述时序跟踪算法可采用现有的时序滤波的目标跟踪方法，例如中国专利CN110111358A一种基于多层时序滤波的目标跟踪方法，此不赘述。
所述人脸匹配模块130用于判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物，并根据判断结果确定目标当前位置。具体地，所述判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物是通过计算所有不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征与待跟踪目标的人脸模板之间的人脸相似度，结合预设的人脸相似阈值进行判断。根据判断结果可确定目标当前位置。所述人脸匹配模块130获取所有不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征，计算所有不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征与待跟踪目标的人脸模板之间的人脸相似度，判断所述人脸相似度是否小于等于预设的人脸相似阈值。其中，若所述人脸相似度大于预设的人脸相似阈值，则时序跟踪目标非正确的待跟踪目标，需获取人脸相似度最大的人脸特征对应的人脸，并将该人脸对应的人物切换为待跟踪目标，获取该人物的人体位置作为目标当前位置。所述人脸模板表示人脸特征模板，由待跟踪目标的人脸图像利用基于深度神经网络的人脸识别模型提取获得。若所述人脸相似度小于等于预设的人脸相似阈值，所述人体交叉判断模块140工作。
其中，利用人脸匹配模块130获取时序跟踪目标的当前位置外的区域的人物可以更为快速和方便地判断出待跟踪目标是否已不在所述时序跟踪目标的当前位置对应区域，以对时序跟踪获得的位置信息进行检验，确保跟踪的准确性。
所述人体交叉判断模块140用于根据所述时序跟踪目标的当前位置与其他人物 的人体位置判断时序跟踪目标是否处于交叉状态。其中，交叉状态是指人物的人体位置对应的区域之间发生重叠的状态。具体地，所述人体交叉判断模块140获取当前帧图像中所有的非时序跟踪目标的人体位置对应的区域，计算其与所述时序跟踪目标的当前位置对应的区域的交并比，获取所述交并比大于预设的交并比阈值的人体位置对应区域所对应的人体位置，并将其作为其他人物的人体位置；根据所述时序跟踪目标的当前位置及所述其他人物的人体位置判断时序跟踪目标是否处于交叉状态。其中，交并比代表交集与并集之间的比值，非时序跟踪目标是指所有人物中除时序跟踪目标外的人物，计算非时序跟踪目标的人物的人体位置对应区域与时序跟踪目标的当前位置对应区域的交并比可初步排除与时序跟踪目标的当前位置不交叉的人物的人体位置，便于交叉状态的判断，交并比越大，交叉面积越大。其他人物是指当前帧图像中所有的非时序跟踪目标的人体位置对应区域与所述时序跟踪目标的当前位置对应区域的交并比大于预设的交并比阈值的人体位置对应区域所对应的人物。若时序跟踪目标不处于交叉状态，则时序跟踪目标的当前位置作为目标当前位置。
在确定待跟踪目标处于所述时序跟踪目标的当前位置对应的区域时，利用人体交叉模块140判断时序跟踪目标的当前位置内的时序跟踪目标的交叉状态，可起到双重保障，以提高跟踪目标的准确性。
所述人体匹配模块150用于判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，并根据判断结果确定目标当前位置。具体地，所述判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物是通过提取与时序跟踪目标发生交叉的其他人物的人体特征，计算所述其他人物的人体特征与待跟踪目标的人体模板之间的人体相似度，结合预设的人体相似阈值进行判断。根据判断结果可确定目标当前位置。其中，若所述人体相似度小于等于预设的人体相似阈值，则时序跟踪目标即为需要跟踪的人物，将时序跟踪目标的当前位置作为目标当前位置。所述人体模板表示人体特征模板，由待跟踪目标的人体图像利用基于深度神经网络的人体识别模型提取获得。若所述人体相似度大于预设的人体相似阈值，获取人体相似度最大的人体特征对应的人体，并将该人体对应的人物切换为待跟踪目标，获取该 人物的人体位置作为目标当前位置。当与时序跟踪目标发生交叉的其他人物的人体相似度大于预设的人体相似阈值，则代表所述时序跟踪目标的当前位置对应的人物并非需要跟踪的人物，需从其他人物的人体位置中获取人体相似度最大的人体特征对应的人体所对应的人物作为待跟踪目标，且其对应的人体位置为目标当前位置。利用人体匹配模块150可更好地对发生交叉状态的时序跟踪目标进行跟踪，提高跟踪的准确性。
所述更新模块160用于将目标当前位置作为待跟踪目标的前一帧位置，将当前帧图像作为前一帧图像。
基于上述设计，工作时，人脸人体检测模块110读取当前帧图像，获取并保存当前帧图像中的所有人物的人脸位置及人体位置；时序跟踪模块120获取前一帧图像及待跟踪目标的前一帧位置，根据所述当前帧图像、前一帧图像及待跟踪目标的前一帧位置，采用时序跟踪算法获取时序跟踪目标的当前位置；人脸匹配模块130获取所有不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征，计算所有不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征与待跟踪目标的人脸模板之间的人脸相似度。判断所述人脸相似度是否小于等于预设的人脸相似阈值，以判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物。其中，若所述人脸相似度大于预设的人脸相似阈值，则在时序跟踪目标的当前位置对应的区域外存在与待跟踪目标的人脸匹配的人物，时序跟踪目标非正确的待跟踪目标，获取人脸相似度最大的人脸特征对应的人脸，并将该人脸对应的人物切换为待跟踪目标，获取该人物的人体位置作为目标当前位置，更新模块160工作，更新模块160将目标当前位置作为待跟踪目标的前一帧位置，将当前帧图像作为前一帧图像；且人脸人体检测模块110、时序跟踪模块120及人脸匹配模块130依序工作；若所述人脸相似度小于等于预设的人脸相似阈值，则在时序跟踪目标的当前位置对应的区域外不存在与待跟踪目标的人脸匹配的人物，人体交叉判断模块140工作；所述人体交叉判断模块140根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态，若时序跟踪目标不处于交叉状态，则时序跟踪目标的当前位置作为目标当前位置，更新模块160工作，且人脸 人体检测模块110、时序跟踪模块120、人脸匹配模块130及人体交叉判断模块140依序工作；若时序跟踪目标处于交叉状态，人体匹配模块150工作；人体匹配模块150提取与时序跟踪目标发生交叉的其他人物的人体特征，计算所述其他人物的人体特征与待跟踪目标的人体模板之间的人体相似度，并根据所述人体相似度确定目标当前位置，以判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，若所述人体相似度大于预设的人体相似阈值，则与时序跟踪目标发生交叉的其他人物中存在与待跟踪目标的人体匹配的人物，时序跟踪目标非正确的待跟踪目标，获取人体相似度最大的人体特征对应的人体，并将该人体对应的人物切换为待跟踪目标，获取该人物的人体位置作为目标当前位置，更新模块160工作，且人脸人体检测模块110、时序跟踪模块120、人脸匹配模块130、人体交叉判断模块140及人体匹配模块150依序工作；若所述人体相似度小于等于预设的人体相似阈值，则将时序跟踪目标的当前位置作为目标当前位置，更新模块160工作，且人脸人体检测模块110、时序跟踪模块120、人脸匹配模块130、人体交叉判断模块140及人体匹配模块150依序工作。
本申请实施例的基于视觉的目标跟踪系统将时序跟踪算法、人脸识别及人体识别结合在一起进行目标跟踪，结合时序跟踪算法获得更可靠的目标位置，降低目标人物发生剧烈姿态变化的可能性，将人脸识别作为目标切换的第一优先级，并通过利用人体识别以避免交叉阶段跟踪到错误的人物，降低当目标人物与其他人物无规律交叉时的跟踪错误，提高目标跟踪的准确性，且鲁棒性高。
图7是本申请另一实施例提供的基于视觉的目标跟踪系统的示意性框图。如图7所示，所述目标跟踪系统20是上述实施例的基础上增加了初始化模块270，所述初始化模块270用于根据跟踪需求，在所有人物中确定待跟踪目标，提取并保存待跟踪目标的人脸模板和人体模板，将待跟踪目标的人体位置作为目标当前位置。所述初始化模块270根据跟踪需求，在所有人物中确定待跟踪目标，获取待跟踪目标的人脸图像和人体图像，利用基于深度神经网络的人脸识别模型结合获得的人脸图像提取待跟踪目标的人脸模板，利用基于深度神经网络的人体识别模型结合获得的人体图像提取待跟踪目标的人体模板，将待跟踪目标的人体 位置作为目标当前位置。
基于上述设计，工作时，人脸人体检测模块210读取当前帧图像，获取并保存当前帧图像中的所有人物的人脸位置及人体位置；初始化模块270根据需求在所有人物中确定待跟踪目标，获取待跟踪目标的人脸图像和人体图像，利用基于深度神经网络的人脸识别模型结合获得的人脸图像提取待跟踪目标的人脸模板，利用基于深度神经网络的人体识别模型结合获得的人体图像提取待跟踪目标的人体模板，将待跟踪目标的人体位置作为目标当前位置；更新模块260将目标当前位置作为待跟踪目标的前一帧位置，将当前帧图像作为前一帧图像；人脸人体检测模块再次读取当前帧图像，获取并保存当前帧图像中的所有人物的人脸位置及人体位置；时序跟踪模块220获取前一帧图像及待跟踪目标的前一帧位置，根据所述当前帧图像、前一帧图像及待跟踪目标的前一帧位置，采用时序跟踪算法获取时序跟踪目标的当前位置；人脸匹配模块250获取所有不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征，计算所有不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征与待跟踪目标的人脸模板之间的人脸相似度。判断所述人脸相似度是否小于等于预设的人脸相似阈值，以判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物。其中，若所述人脸相似度大于预设的人脸相似阈值，则在时序跟踪目标的当前位置对应的区域外存在与待跟踪目标的人脸匹配的人物，时序跟踪目标并非需要跟踪的人物，获取人脸相似度最大的人脸特征对应的人脸，并将该人脸对应的人物切换为待跟踪目标，获取该人物的人体位置作为目标当前位置，更新模块260工作，且人脸人体检测模块210、时序跟踪模块220及人脸匹配模块250依序工作；若所述人脸相似度小于等于预设的人脸相似阈值，人体交叉判断模块240工作；所述人体交叉判断模块240根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态，若时序跟踪目标不处于交叉状态，则时序跟踪目标的当前位置作为目标当前位置，更新模块260工作，且人脸人体检测模块210、时序跟踪模块220、人脸匹配模块250及人体交叉判断模块240依序工作；若时序跟踪目标处于交叉状态，人体匹配模块250工作；人体匹配模块250提取与时序跟踪目标发生交叉的其他人 物的人体特征，计算所述其他人物的人体特征与待跟踪目标的人体模板之间的人体相似度，并根据所述人体相似度确定目标当前位置，以判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物。若所述人体相似度大于预设的人体相似阈值，则与时序跟踪目标发生交叉的其他人物中存在与待跟踪目标的人体匹配的人物，时序跟踪目标并非需要跟踪的人物，获取人体相似度最大的人体特征对应的人体，并将该人体对应的人物切换为待跟踪目标，获取该人物的人体位置作为目标当前位置，更新模块260工作，且人脸人体检测模块210、时序跟踪模块220、人脸匹配模块230、人体交叉判断模块240及人体匹配模块250依序工作；若所述人体相似度小于等于预设的人体相似阈值，则将时序跟踪目标的当前位置作为目标当前位置，更新模块260工作，且人脸人体检测模块210、时序跟踪模块220、人脸匹配模块230、人体交叉判断模块240及人体匹配模块250依序工作。
本申请实施例通过设置时序跟踪模块、人脸匹配模块、人体交叉判断模块及人体匹配模块以将时序跟踪算法、人脸识别及人体识别多种目标跟踪方式结合在一起进行目标跟踪，结合时序跟踪算法以利用待跟踪目标的时序信息获得更可靠的目标位置，降低目标人物发生剧烈姿态变化的可能性，可在待跟踪目标不处于交叉状态时获取可靠的目标位置；将人脸识别作为目标切换的第一优先级，以对时序跟踪目标进行修正，并通过利用人体识别以有效避免交叉阶段跟踪到错误的人物，降低当目标人物与其他人物无规律交叉时的跟踪错误，提高目标跟踪的准确性，且鲁棒性高。
上述基于视觉的目标跟踪系统可以实现为一种计算机程序的形式，该计算机程序可以在如图8所示的计算机设备上运行。
请参阅图8，图8是本申请实施例的一种计算机设备的示意性框图。该计算机设备30可以是终端。也可以是服务器，其中，终端可以是平板电脑、笔记本电脑和台式电脑等具有通信功能的电子设备。服务器可以是独立的服务器，也可以是多个服务器组成的服务器集群。
参阅图8，该计算机设备30包括通过系统总线301连接的处理器302、存储器和网络接口305，其中，存储器可以包括非易失性存储介质303和内存储器304。该 非易失性存储介质303可存储操作系统3031和计算机程序3032。该计算机程序3032包括程序指令，该程序指令被执行时，可使得处理器302执行一种基于视觉的目标跟踪方法。该处理器302用于提供计算和控制能力，以支撑整个计算机设备30的运行。该内存储器304为非易失性存储介质303中的计算机程序3032的运行提供环境，该计算机程序3032被处理器302执行时，可使得处理器302执行一种基于视觉的目标跟踪方法。该网络接口305用于与其它设备进行网络通信。本领域技术人员可以理解，图8中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备30的限定，具体的计算机设备30可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
其中，所述处理器302用于运行存储在存储器中的计算机程序3032，以实现一种基于视觉的目标跟踪方法，该目标跟踪方法包括：步骤A、读取当前帧图像，获取并保存当前帧图像的所有人物的人脸位置及人体位置；步骤B、获取前一帧图像及待跟踪目标的前一帧位置，根据所述当前帧图像、前一帧图像及待跟踪目标的前一帧位置，采用时序跟踪算法获取时序跟踪目标的当前位置；步骤C、判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物，若不存在，执行步骤D；步骤D、根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态，若是，执行步骤E；步骤E、判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，若不存在，则将时序跟踪目标的当前位置作为目标当前位置，执行步骤F；步骤F、保持跟踪，执行步骤G；步骤G、将目标当前位置作为待跟踪目标的前一帧位置，将当前帧图像作为前一帧图像，返回执行步骤A。
本申请实施例所提供的一种计算机设备，其内存储的计算机程序不限于如上的方法操作，还可以执行本申请任意实施例所提供的基于视觉的目标跟踪方法中的相关操作。
应当理解，在本申请实施例中，处理器302可以是中央处理单元(Central Processing Unit，CPU)，该处理器302还可以是其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated  Circuit，ASIC)、现成可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。其中，通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
本领域普通技术人员可以理解的是实现上述实施例的方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成。该计算机程序包括程序指令，计算机程序可存储于一存储介质中，该存储介质为计算机可读存储介质。该程序指令被该计算机系统中的至少一个处理器执行，以实现上述方法的实施例的流程步骤。
因此，本申请还提供一种存储介质。该存储介质可以为计算机可读存储介质。该存储介质存储有计算机程序，其中计算机程序包括程序指令。该程序指令被处理器执行时使处理器实现一种基于视觉的目标跟踪方法，该目标跟踪方法包括：步骤A、读取当前帧图像，获取并保存当前帧图像的所有人物的人脸位置及人体位置；步骤B、获取前一帧图像及待跟踪目标的前一帧位置，根据所述当前帧图像、前一帧图像及待跟踪目标的前一帧位置，采用时序跟踪算法获取时序跟踪目标的当前位置；步骤C、判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物，若不存在，执行步骤D；步骤D、根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态，若是，执行步骤E；步骤E、判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，若不存在，则将时序跟踪目标的当前位置作为目标当前位置，执行步骤F；步骤F、保持跟踪，执行步骤G；步骤G、将目标当前位置作为待跟踪目标的前一帧位置，将当前帧图像作为前一帧图像，返回执行步骤A。
本申请实施例所提供的一种存储介质，其内存储的计算机程序包括的程序指令不限于如上的方法操作，还可以执行本申请任意实施例所提供的基于视觉的目标跟踪方法中的相关操作。
所述存储介质可以是移动硬盘、只读存储器(Read-Only Memory，ROM)、磁碟或者光盘等各种可以存储程序代码的计算机可读存储介质。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的 单元及算法步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。技术人员可对每个特定的应用使用不同方法来实现所描述的功能，但这种实现不应认为超出本申请的范围。在本申请所提供的几个实施例中，应该理解到，所揭露的系统和方法，可通过其它的方式实现。本申请实施例方法中的步骤可以根据实际需要进行顺序调整、合并和删减。本申请实施例系统中的单元可以根据实际需要进行合并、划分和删减。另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以是两个或两个以上单元集成在一个单元中。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到各种等效的修改或替换，这些修改或替换都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以权利要求的保护范围为准。
发明概述
技术问题
问题的解决方案
发明的有益效果

Claims (10)

  1. 一种基于视觉的目标跟踪方法，其特征在于，包括：
    步骤A、读取当前帧图像，获取并保存当前帧图像中的所有人物的人脸位置及人体位置；
    步骤B、获取前一帧图像及待跟踪目标的前一帧位置，根据所述当前帧图像、前一帧图像及待跟踪目标的前一帧位置，采用时序跟踪算法获取时序跟踪目标的当前位置；
    步骤C、判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物，若不存在，执行步骤D；
    步骤D、根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态，若是，执行步骤E；
    步骤E、判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，若不存在，将时序跟踪目标的当前位置作为目标当前位置，执行步骤F；
    步骤F、保持跟踪，执行步骤G；
    步骤G、将目标当前位置作为待跟踪目标的前一帧位置，将当前帧图像作为前一帧图像，返回执行步骤A。
  2. 如权利要求1所述的基于视觉的目标跟踪方法，其特征在于，所述步骤C具体包括：
    步骤C1、获取所有不在所述时序跟踪目标的当前位置对应的区域内的人物对应的人脸特征，计算其与待跟踪目标的人脸模板之间的人脸相似度；
    步骤C2、判断所述人脸相似度是否小于等于预设的人脸相似阈值，若是，执行步骤D。
  3. 如权利要求1所述的基于视觉的目标跟踪方法，其特征在于，所述步骤C还包括：
    步骤c、若在时序跟踪目标的当前位置对应的区域外存在与待跟踪目标的人脸匹配的人物，将该人物切换为待跟踪目标，获取该人 物的人体位置作为目标当前位置，执行步骤G。
  4. 如权利要求1所述的基于视觉的目标跟踪方法，其特征在于，所述步骤D还包括：
    步骤d、若时序跟踪目标不处于交叉状态，则将时序跟踪目标的当前位置作为目标当前位置，执行步骤G。
  5. 如权利要求1所述的基于视觉的目标跟踪方法，其特征在于，所述步骤E包括：
    步骤E1、提取与时序跟踪目标发生交叉的其他人物的人体特征，计算所述其他人物的人体特征与待跟踪目标的人体模板之间的人体相似度；
    步骤E2、判断所述人体相似度是否小于等于预设的人体相似阈值；
    步骤E3、若是，将时序跟踪目标的当前位置作为目标当前位置，执行步骤F。
  6. 如权利要求1所述的基于视觉的目标跟踪方法，其特征在于，所述步骤E还包括：
    步骤e、若与时序跟踪目标发生交叉的其他人物中存在与待跟踪目标的人体匹配的人物，将该人物切换为待跟踪目标，获取该人物的人体位置作为目标当前位置，执行步骤G。
  7. 如权利要求1所述的基于视觉的目标跟踪方法，其特征在于，所述步骤B前还具体包括：
    步骤a1、判断是否已确定待跟踪目标，若是，执行步骤B，否则，执行步骤a2；
    步骤a2、根据跟踪需求，在所有人物中确定待跟踪目标，提取并保存待跟踪目标的人脸模板和人体模板；
    步骤a3、将待跟踪目标的人体位置作为目标当前位置，执行步骤G。
  8. 一种基于视觉的目标跟踪系统，其特征在于，包括：
    人脸人体检测模块，用于读取当前帧图像，获取并保存当前帧图像中的所有人物的人脸位置及人体位置；
    时序跟踪模块，用于获取前一帧图像及待跟踪目标的前一帧位置，根据所述当前帧图像、前一帧图像及待跟踪目标的前一帧位置，采用时序跟踪算法获取时序跟踪目标的当前位置；
    人脸匹配模块，用于判断在时序跟踪目标的当前位置对应的区域外是否存在与待跟踪目标的人脸匹配的人物，并根据判断结果确定目标当前位置；
    人体交叉判断模块，用于根据所述时序跟踪目标的当前位置与其他人物的人体位置判断时序跟踪目标是否处于交叉状态；
    人体匹配模块，用于判断与时序跟踪目标发生交叉的其他人物中是否存在与待跟踪目标的人体匹配的人物，并根据判断结果确定目标当前位置；
    更新模块，用于将目标当前位置作为待跟踪目标的前一帧位置，将当前帧图像作为前一帧图像。
  9. 一种计算机设备，其特征在于，所述计算机设备包括存储器和处理器，所述存储器上存储有可在处理器运行的计算机程序，所述处理器执行所述计算机程序时实现如权利要求1-7任意一项所述的基于视觉的目标跟踪方法。
  10. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质存储有计算机程序，所述计算机程序包括程序指令，所述程序指令当被执行时实现如权利要求1-7任意一项所述的基于视觉的目标跟踪方法。
PCT/CN2020/076360 2020-02-14 2020-02-24 基于视觉的目标跟踪方法、系统、设备及存储介质 WO2021159554A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/886,515 US20220383518A1 (en) 2020-02-14 2022-08-12 Vision-based target tracking method, system, equipment, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010095228.9 2020-02-14
CN202010095228.9A CN111339855B (zh) 2020-02-14 2020-02-14 基于视觉的目标跟踪方法、系统、设备及存储介质

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/886,515 Continuation US20220383518A1 (en) 2020-02-14 2022-08-12 Vision-based target tracking method, system, equipment, and storage medium

Publications (1)

Publication Number Publication Date
WO2021159554A1 true WO2021159554A1 (zh) 2021-08-19

Family

ID=71181571

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/076360 WO2021159554A1 (zh) 2020-02-14 2020-02-24 基于视觉的目标跟踪方法、系统、设备及存储介质

Country Status (3)

Country Link
US (1) US20220383518A1 (zh)
CN (1) CN111339855B (zh)
WO (1) WO2021159554A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116168421A (zh) * 2023-04-13 2023-05-26 微网优联科技(成都)有限公司 一种基于ipc网络摄像机的目标追踪方法及系统

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111815676A (zh) * 2020-07-06 2020-10-23 浙江大华技术股份有限公司 人脸跟踪方法、装置、设备及存储装置
CN111770312A (zh) * 2020-07-13 2020-10-13 深圳市元黎技术有限公司 一种安防监控方法及安防监控系统
CN112069879B (zh) * 2020-07-22 2024-06-07 深圳市优必选科技股份有限公司 一种目标人物跟随方法、计算机可读存储介质及机器人
JP2022059972A (ja) * 2020-10-02 2022-04-14 本田技研工業株式会社 認識対象者の認識方法
CN112287846B (zh) * 2020-10-30 2024-05-07 深圳市优必选科技股份有限公司 目标人物跟随方法、装置、可移动机器人及可读存储介质
CN114449155A (zh) * 2020-11-02 2022-05-06 嘉楠明芯(北京)科技有限公司 基于摄像装置的云台控制方法以及云台控制装置
CN112614168B (zh) * 2020-12-21 2023-08-29 浙江大华技术股份有限公司 一种目标人脸跟踪方法、装置、电子设备及存储介质
CN112907652B (zh) * 2021-01-25 2024-02-02 脸萌有限公司 相机姿态获取方法、视频处理方法、显示设备和存储介质
CN113642454B (zh) * 2021-08-11 2024-03-01 汇纳科技股份有限公司 座椅使用情况的识别方法、系统、设备及计算机存储介质
CN114241586B (zh) * 2022-02-21 2022-05-27 飞狐信息技术（天津）有限公司 一种人脸检测方法及装置、存储介质及电子设备
CN114779271B (zh) * 2022-06-16 2022-10-21 杭州宏景智驾科技有限公司 目标检测方法和装置、电子设备和存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080187175A1 (en) * 2007-02-07 2008-08-07 Samsung Electronics Co., Ltd. Method and apparatus for tracking object, and method and apparatus for calculating object pose information
CN102057665A (zh) * 2009-05-07 2011-05-11 松下电器产业株式会社 电子照相机、图象处理装置及图象处理方法
CN104346811A (zh) * 2014-09-30 2015-02-11 深圳市华尊科技有限公司 基于视频图像的目标实时追踪方法及其装置
CN107316322A (zh) * 2017-06-27 2017-11-03 上海智臻智能网络科技股份有限公司 视频跟踪方法和装置、以及对象识别方法和装置
CN110414443A (zh) * 2019-07-31 2019-11-05 苏州市科远软件技术开发有限公司 一种目标跟踪方法、装置及枪球联动跟踪方法

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2009048347A (ja) * 2007-08-17 2009-03-05 Seiko Epson Corp 画像処理装置、画像処理方法、および、画像処理プログラム
JP2012226403A (ja) * 2011-04-15 2012-11-15 Nec Corp 画像領域追跡装置、画像領域追跡方法、および、コンピュータ・プログラム
CN103310462B (zh) * 2013-05-22 2016-07-06 成都信通易安科技有限公司 一种基于联合特征模型的多人体跟踪方法
CN107644204B (zh) * 2017-09-12 2020-11-10 南京凌深信息科技有限公司 一种用于安防系统的人体识别与跟踪方法
CN109446942B (zh) * 2018-10-12 2020-10-16 北京旷视科技有限公司 目标跟踪方法、装置和系统
CN109934098A (zh) * 2019-01-24 2019-06-25 西北工业大学 一种带隐私保护的摄像机智能系统及其实现方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080187175A1 (en) * 2007-02-07 2008-08-07 Samsung Electronics Co., Ltd. Method and apparatus for tracking object, and method and apparatus for calculating object pose information
CN102057665A (zh) * 2009-05-07 2011-05-11 松下电器产业株式会社 电子照相机、图象处理装置及图象处理方法
CN104346811A (zh) * 2014-09-30 2015-02-11 深圳市华尊科技有限公司 基于视频图像的目标实时追踪方法及其装置
CN107316322A (zh) * 2017-06-27 2017-11-03 上海智臻智能网络科技股份有限公司 视频跟踪方法和装置、以及对象识别方法和装置
CN110414443A (zh) * 2019-07-31 2019-11-05 苏州市科远软件技术开发有限公司 一种目标跟踪方法、装置及枪球联动跟踪方法

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116168421A (zh) * 2023-04-13 2023-05-26 微网优联科技(成都)有限公司 一种基于ipc网络摄像机的目标追踪方法及系统

Also Published As

Publication number Publication date
US20220383518A1 (en) 2022-12-01
CN111339855A (zh) 2020-06-26
CN111339855B (zh) 2023-05-23

Similar Documents

Publication Publication Date Title
WO2021159554A1 (zh) 基于视觉的目标跟踪方法、系统、设备及存储介质
WO2020199479A1 (zh) 一种人体动作的识别方法及设备
WO2020199480A1 (zh) 一种人体动作的识别方法及设备
JP7342919B2 (ja) 情報処理システム、制御方法、及びプログラム
WO2019184749A1 (zh) 轨迹跟踪方法、装置、计算机设备和存储介质
CN109977906B (zh) 手势识别方法及系统、计算机设备及存储介质
JP7260674B2 (ja) ニューラルネットワークに基づくｃ／ｄ比決定方法、コンピュータ機器及び記憶媒体
KR102432204B1 (ko) 이미지 처리 방법, 장치, 저장 매체 및 전자 기기
KR102349059B1 (ko) 영상의 관심영역으로부터 랜드마크를 결정하는 장치 및 방법
US11354883B2 (en) Image processing method and apparatus, and electronic device
JP2021064367A (ja) 動作認識方法、動作認識装置及び電子機器
CN111754391A (zh) 人脸转正方法、设备及计算机可读存储介质
KR20220004009A (ko) 키 포인트 검출 방법, 장치, 전자 기기 및 저장 매체
CN113822163A (zh) 一种复杂场景下的行人目标跟踪方法及装置
WO2020137069A1 (ja) 位置推定システム
JP7065557B2 (ja) 人物を追跡する映像解析装置、プログラム及び方法
CN116645396A (zh) 轨迹确定方法、装置、计算机可读存储介质及电子设备
WO2021051868A1 (zh) 目标的定位方法及装置、计算机设备、计算机存储介质
AU2020294190B2 (en) Image processing method and apparatus, and electronic device
JP2023155187A (ja) レーダ点群に基づく目標追跡方法及び装置
JP5253227B2 (ja) 画像入力装置及び被写体検出方法、プログラム
JP7115579B2 (ja) 情報処理装置、情報処理方法、及びプログラム
US20100014710A1 (en) Method and system for tracking positions of human extremities
CN111401285B (zh) 目标跟踪方法、装置及电子设备
JP2022140105A (ja) 物体追跡装置および物体追跡方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20918170

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 04.01.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20918170

Country of ref document: EP

Kind code of ref document: A1