CN112200076A - Method for carrying out multi-target tracking based on head and trunk characteristics - Google Patents

Method for carrying out multi-target tracking based on head and trunk characteristics Download PDF

Info

Publication number
CN112200076A
CN112200076A CN202011076008.8A CN202011076008A CN112200076A CN 112200076 A CN112200076 A CN 112200076A CN 202011076008 A CN202011076008 A CN 202011076008A CN 112200076 A CN112200076 A CN 112200076A
Authority
CN
China
Prior art keywords
image
head
target
human body
key points
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
Application number
CN202011076008.8A
Other languages
Chinese (zh)
Other versions
CN112200076B (en
Inventor
柯逍
叶宇
李悦洲
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fuzhou University
Original Assignee
Fuzhou University
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 Fuzhou University filed Critical Fuzhou University
Priority to CN202011076008.8A priority Critical patent/CN112200076B/en
Publication of CN112200076A publication Critical patent/CN112200076A/en
Application granted granted Critical
Publication of CN112200076B publication Critical patent/CN112200076B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Closed-Circuit Television Systems (AREA)
  • Image Analysis (AREA)

Abstract

本发明涉及一种基于头部躯干特征进行多目标跟踪的方法，包括以下步骤：步骤S1：获取视频中行人检测的结果，并对结果进行筛选，删除错误的检测结果；步骤S2：对筛选后的检测结果进行预处理后，输入到人体关键点检测网络中，得到所有的人体关键点；步骤S3：对得到的每个行人的关键点进行筛选，选取头部以及肩膀的关键点进行组合，获得头部躯干特征；步骤S4：将得到的单个行人的头部躯干特征输入到跟踪器中进行初始化，进而对目标进行跟踪。本发明能够有效地提取检测框中的头部躯干特征，使得跟踪器在初始化的时候获取的有效信息占比最大化。

Figure 202011076008

The present invention relates to a method for multi-target tracking based on head and torso features, comprising the following steps: Step S1 : obtaining pedestrian detection results in videos, screening the results, and deleting wrong detection results; Step S2 : after screening After preprocessing the detected results, input them into the human key point detection network to obtain all human key points; Step S3: Screen the obtained key points of each pedestrian, and select the key points of the head and shoulder to combine, Obtain head and torso features; Step S4: Input the obtained head and torso features of a single pedestrian into the tracker for initialization, and then track the target. The invention can effectively extract the features of the head and torso in the detection frame, so that the proportion of the effective information obtained by the tracker during initialization is maximized.

Figure 202011076008

Description

Method for carrying out multi-target tracking based on head and trunk characteristics
Technical Field
The invention relates to the field of computer vision, in particular to a method for multi-target tracking based on head and trunk characteristics.
Background
Multi-Object Tracking (MOT). The main task is to give an image sequence, find moving objects in the image sequence, correspond moving objects in different frames one to one (Identity), and then give the motion tracks of different objects. The mainstream framework adopted by the academia in the multi-target Tracking (MOT) problem at present is TBD (Tracking-by-Detection), that is, Tracking based on Detection, and in this mainstream Tracking framework, the multi-target Tracking problem is expressed as an association matching problem: if the detection result obtained from a certain frame is matched with the detection result obtained from the previous frame, the same target is identified.
With the continuous development of single-target trackers in recent years, a large number of trackers with good tracking effect and high running speed appear. In previous work, a single-target tracker has been applied to a multi-target tracking task and achieved certain effect, but the performance of the single-target tracker in a complex scene (such as an MOT19 data set) is not ideal, because in the complex scene, a large amount of redundant features and interference information are contained in a detection frame, and the tracking effect is greatly affected by initializing the tracker with the features and the information.
Disclosure of Invention
In view of this, the present invention provides a method for performing multi-target tracking based on head and torso features, which can effectively extract head and torso features in a detection frame, so as to maximize an effective information ratio obtained by a tracker during initialization.
In order to achieve the purpose, the invention adopts the following technical scheme:
a method for multi-target tracking based on head and trunk characteristics comprises the following steps:
step S1: acquiring a pedestrian detection result in the video, screening the result, and deleting an error detection result;
step S2: preprocessing the screened detection results, and inputting the preprocessed detection results into a human body key point detection network to obtain all human body key points;
step S3: screening the obtained key points of each pedestrian, selecting the key points of the head and the shoulders to combine to obtain the characteristics of the head and the trunk;
step S4: and inputting the obtained head and trunk characteristics of the single pedestrian into a tracker for initialization, and further tracking the target.
Further, the step S1 is specifically:
step S11, after each frame of image of the video is preprocessed, the image is detected by using a target detection network;
step S12: detecting information of each frame of the video by using a pedestrian detector to obtain a detection result R, and enabling R to be { K ═ Ki,PjDet _ x, det _ y, det _ w, det _ h, det _ c, i 1,2,. M, j 1,2,. N, representing the set of all detection results in a video sequence, where M represents the number of all image frames in a video sequence, N represents the number of all detected pedestrians in a frame of image, and K represents the number of detected pedestrians in a frame of imageiRepresenting the ith frame of a video sequence, PjRepresenting the jth pedestrian in the frame image, det _ x, det _ y, det _ w and det _ h respectively represent the x coordinate and the y coordinate of the upper left corner of the detection frame of the pedestrian and the width and the height of the detection frame, and det _ c represents the confidence coefficient of the detection frame;
step S13: let the confidence threshold value of pedestrian detection be TdThe pedestrian aspect ratio threshold is TrDeleting the detection results satisfying the following conditions:
det_c＜Td or det_w/det_h＞Tr。
further, the step S2 is specifically:
step S21, the screened detection frame resize is set to a preset size;
step S22: preprocessing the image after resize, copying the preprocessed image, horizontally turning the copied image, and inputting the original image and the turned image into a human body key point detection network;
step S23: obtaining the output result S of the human body key point detection network, and enabling S to be { J ═ JzZ, where Z represents the number of human keypoints in the image; j. the design is a squarez- { jo int _ x, jo int _ y, jo int _ c } denotes the thz key points, wherein jo int _ x represents the x coordinate of the human body key point, jo int _ y represents the y coordinate of the human body key point, and jo int _ c represents the confidence coefficient of the human body key point;
step S24: let the human body key point detection result of the original image be SsrcLet the human body key point detection result of the flip image be SflipAnd fusing the two detection results.
Further, the pretreatment specifically comprises: firstly, removing sharp noise of an image by using Gaussian filtering, and then removing fine interference details by using a USM sharpening enhancement algorithm, wherein the calculation method comprises the following steps:
Figure BDA0002716672720000031
where output represents an output image, orign _ image represents an original image, gaus _ image represents an image after gaussian filtering, and ω represents a USM coefficient.
Further, the fusion method comprises the following steps:
Figure BDA0002716672720000032
Figure BDA0002716672720000033
Figure BDA0002716672720000034
wherein c issrcRepresenting confidence, x, of key points of the body in the original imagesrcX-coordinate, y representing key points of a human body in an original imagesrcY-coordinate representing key points of the human body in the original image, where cflipRepresenting confidence, x, of key points of a human body in a flip imageflipX-coordinate, y representing key points of a human body in a flip imageflipY-coordinate representing key points of the human body in the flip image, final _ x and final _ y, respectivelyAnd x and y coordinates of the human key points of which the final fusion is finished are represented, and final _ c represents the confidence coefficient of the human key points of which the final fusion is finished.
Further, the step S3 is specifically:
step S31: screening the selected key points, and adopting a screening scheme based on confidence coefficient to enable the detection confidence coefficient threshold of the human key points to be TkpDeleting the human body key points meeting the following requirements:
jo int_c＜Tkp
step S32: combining the screened key points, enabling the screened human body key point set to be Q, traversing the set Q, sequencing the whole set Q according to the sizes of x coordinates and y coordinates, finding out the human body key points at the top, the bottom, the left and the right in the set, and obtaining the minimum rectangular convex hull in the set, wherein the image content in the rectangular convex hull is the head and trunk characteristics of the target.
Further, the step S4 is specifically:
step S41: let the set of targets in the tracked state be OtrackThe set comprises all targets in a tracking state from a first frame of the video to a current frame;
step S42: traversing a set of targets O in a state being trackedtrackPerforming IOU and OKS calculation on the newly obtained head and trunk characteristics and all targets in the set to confirm whether the target is in a tracking state;
step S43: calculating the fusion metric value of the bounding box of the tracked target and the newly obtained head and trunk characteristic bounding box in the target set in the tracked state, wherein the calculation method is as follows
Figure BDA0002716672720000041
If FF is greater than 0.5, the target is considered to exist, and the target does not need to be initialized to a new target again, and the step S1 is executed; otherwise the feature is identified as belonging to a new target, proceeding to the next step;
step S44: inputting new head and trunk characteristics into a tracker for initialization, and adding a target set O in a tracked statetrackGo to step S1.
Further, the method for calculating the IOU and OKS is as follows:
Figure BDA0002716672720000042
Figure BDA0002716672720000043
wherein A represents the area of the bounding box of the first object and B represents the area of the bounding box of the second object; the calculation method of the bounding box area is that the length of the rectangle is multiplied by the width of the rectangle; viszIndicates the visibility of the z-th keypoint (greater than 0 indicating visibility), dis indicates the Euclidean distance, scale, of the existing and detected human keypoints2The square root, σ, representing the size of the area occupied by these key pointszA normalization factor representing the z-th human body key point.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention does not depend on the quality of the detection result, and can also be corrected through the human body key point detection network even if the detection frame has deviation with the group Truth in the data set;
2. according to the invention, the head trunk characteristics acquired through the human body key point detection network are used for target tracking, and the angle of monitoring scene shooting is considered, so that the head trunk is not easy to be shielded, and effective characteristics can be extracted for tracking even in a monitoring scene crowded with people;
3. the method adopts the human body key point detection network to obtain effective information contained in the head and trunk characteristics, and the effective information has larger average proportion in the image, thereby being more beneficial to the initialization and the subsequent tracking of the tracker.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
The invention is further explained below with reference to the drawings and the embodiments.
Referring to fig. 1, the present invention provides a method for multi-target tracking based on head and torso features, comprising the following steps:
step S1: acquiring a pedestrian detection result in the video, screening the result, and deleting an error detection result;
step S2: inputting the screened detection result into a human body key point detection network to obtain all human body key points; the dependence degree on the quality of the detection result can be reduced, and even if the quality of the detection result is poor, the detection result can be corrected through key points of a human body;
step S3: considering the shooting angle of a monitoring scene, as the head trunk is not easy to be shielded, effective features can be extracted for tracking even in the monitoring scene crowded with people, the obtained key points of each pedestrian are screened, and then the key points of the head and the shoulders are selected for combination to obtain the head trunk features;
step S4: the obtained head and trunk characteristics of the single pedestrian are input into the tracker to be initialized, and then the target is tracked.
In this embodiment, the step S1 includes the following steps:
step S11: after each frame of image of the video is preprocessed, a target detection network is utilized
Detecting the same;
step S12: detecting information of each frame of the video by using a pedestrian detector to obtain a detection result R, and enabling R to be { K ═ Ki,PjDet _ x, det _ y, det _ w, det _ h, det _ c, i 1,2,. M, j 1,2,. N, representing the set of all detection results in a video sequence, where M represents all images in a video sequenceNumber of frames, N represents the number of all detected pedestrians in one frame image, KiRepresenting the ith frame of a video sequence, PjRepresenting the jth pedestrian in the frame image, det _ x, det _ y, det _ w and det _ h respectively represent the x coordinate and the y coordinate of the upper left corner of the detection frame of the pedestrian and the width and the height of the detection frame, and det _ c represents the confidence coefficient of the detection frame;
step S13: let the confidence threshold value of pedestrian detection be TdThe pedestrian aspect ratio threshold is TrDeleting the detection results satisfying the following conditions:
det_c＜Td or det_w/det_h＞Tr
in this embodiment, step S2 specifically includes the following steps:
step S21: the screened detection box resize is to a size of 4:3, using a size of 344 × 258;
step S22: preprocessing the image after resize, firstly removing sharp noise of the image by using Gaussian filtering, wherein the Gaussian filtering is used because boundary information can be better reserved; then, a USM sharpening enhancement algorithm is used for removing fine interference details, and the calculation method is as follows:
Figure BDA0002716672720000061
where output represents an output image, orign _ image represents an original image, gaus _ image represents an image after gaussian filtering, and ω represents a USM coefficient. Copying the preprocessed output image, horizontally turning the copied image, and inputting the original image and the turned image into a human body key point detection network;
step S23: obtaining the output result S of the human body key point detection network, and enabling S to be { J ═ JzZ, where Z represents the number of human keypoints in the image. J. the design is a squarezThe z-th key point is represented by { jo int _ x, jo int _ y and jo int _ c }, wherein the jo int _ x represents the x coordinate of the human body key point, the jo int _ y represents the y coordinate of the human body key point, and the jo int _ c represents the confidence coefficient of the human body key point;
step S24: let the human body key point detection result of the original image be SsrcLet the human body key point detection result of the flip image be SflipAnd fusing the two detection results, wherein the step is to achieve more accurate human body key point coordinates, and the fusing method comprises the following steps:
Figure BDA0002716672720000071
Figure BDA0002716672720000072
Figure BDA0002716672720000073
wherein c issrcRepresenting confidence, x, of key points of the body in the original imagesrcX-coordinate, y representing key points of a human body in an original imagesrcY-coordinate representing key points of the human body in the original image, where cflipRepresenting confidence, x, of key points of a human body in a flip imageflipX-coordinate, y representing key points of a human body in a flip imageflipAnd the y coordinate represents the human body key point in the overturned image, final _ x and final _ y represent the x and y coordinates of the human body key point after final fusion is finished respectively, and final _ c represents the confidence coefficient of the human body key point after final fusion is finished.
In this embodiment, selecting human body key points with good tracking characteristics specifically includes: human body key points of the head (eyes, nose, ears) and two shoulders, the step S3 specifically includes the following steps:
step S31: screening the selected key points, and adopting a screening scheme based on confidence coefficient to enable the detection confidence coefficient threshold of the human key points to be TkpDeleting the human body key points meeting the following requirements:
jo int_c＜Tkp
step S32: combining the screened key points, enabling the screened human body key point set to be Q, traversing the set Q, sorting the whole set Q according to the sizes of x coordinates and y coordinates respectively, finding out the human body key points at the top, the bottom, the left and the right in the set, and obtaining the minimum rectangular convex hull in the set. The image content within the rectangular convex hull is the head-torso feature of the target.
In this embodiment, step S4 specifically includes the following steps:
step S41: let the set of targets in the tracked state be OtrackThe set contains all the objects in the tracking state from the first frame of the video to the current frame.
Step S42: traversing a set of targets O in a state being trackedtrackAnd performing IOU (cross-over ratio) and OKS (object key point similarity) calculation on the newly obtained head and trunk features and all the objects in the set to confirm whether the object is in a tracking state or not, wherein the IOU and OKS calculation method comprises the following steps:
Figure BDA0002716672720000081
Figure BDA0002716672720000082
wherein A represents the area of the bounding box of the first object and B represents the area of the bounding box of the second object; the calculation method of the bounding box area is that the length of the rectangle is multiplied by the width of the rectangle; viszIndicates the visibility of the z-th keypoint (greater than 0 indicating visibility), dis indicates the Euclidean distance, scale, of the existing and detected human keypoints2The square root, σ, representing the size of the area occupied by these key pointszAnd (3) a normalization factor representing the z-th human body key point (the factor is obtained by calculating the standard deviation of all group Truth in the existing data set, and reflects the influence degree of the current key point on the whole body).
Step S43: calculating the fusion metric value of the bounding box of the tracked target and the newly obtained head and trunk characteristic bounding box in the target set in the tracked state, wherein the calculation method is as follows
Figure BDA0002716672720000091
If FF is greater than 0.5, the target is considered to exist, and the target does not need to be initialized to a new target again, and the step S11 is executed; otherwise the feature is deemed to be subject to a new target and the next step is taken.
Step S44: inputting new head and trunk characteristics into a tracker for initialization, and adding a target set O in a tracked statetrackGo to step S11.
The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.

Claims (8)

1. A method for multi-target tracking based on head and trunk characteristics is characterized by comprising the following steps:
step S1: acquiring a pedestrian detection result in the video, screening the result, and deleting an error detection result;
step S2: preprocessing the screened detection results, and inputting the preprocessed detection results into a human body key point detection network to obtain all human body key points;
step S3: screening the obtained key points of each pedestrian, selecting the key points of the head and the shoulders to combine to obtain the characteristics of the head and the trunk;
step S4: and inputting the obtained head and trunk characteristics of the single pedestrian into a tracker for initialization, and further tracking the target.
2. The method for multi-target tracking based on head-torso features of claim 1, wherein the step S1 is specifically performed by:
step S11, after each frame of image of the video is preprocessed, the image is detected by using a target detection network;
step S12: detecting information of each frame of the video by using a pedestrian detector to obtain a detection result R, and enabling R to be { K ═ Ki,PjDet _ x, det _ y, det _ w, det _ h, det _ c, i 1,2,. M, j 1,2,. N, representing the set of all detection results in a video sequence, where M represents the number of all image frames in a video sequence, N represents the number of all detected pedestrians in a frame of image, and K represents the number of detected pedestrians in a frame of imageiRepresenting the ith frame of a video sequence, PjRepresenting the jth pedestrian in the frame image, det _ x, det _ y, det _ w and det _ h respectively represent the x coordinate and the y coordinate of the upper left corner of the detection frame of the pedestrian and the width and the height of the detection frame, and det _ c represents the confidence coefficient of the detection frame;
step S13: let the confidence threshold value of pedestrian detection be TdThe pedestrian aspect ratio threshold is TrDeleting the detection results satisfying the following conditions:
det_c＜Td or det_w/det_h＞Tr。
3. the method for multi-target tracking based on head-torso features of claim 1, wherein the step S2 is specifically performed by:
step S21, the screened detection frame resize is set to a preset size;
step S22: preprocessing the image after resize, copying the preprocessed image, horizontally turning the copied image, and inputting the original image and the turned image into a human body key point detection network;
step S23: obtaining the output result S of the human body key point detection network, and enabling S to be { J ═ JzZ, where Z represents the number of human keypoints in the image; j. the design is a squarez-join _ x, join _ y, join _ c represents the z-th keypoint, where join _ x represents the x-coordinate of a human keypoint, join _ y represents the y-coordinate of a human keypoint, and join _ c represents the confidence of this human keypoint;
step S24: let the human body key point detection result of the original image be SsrcTo turn over the human body of the imageThe key point detection result is SflipAnd fusing the two detection results.
4. The method for multi-target tracking based on head and torso features of claim 3, wherein the preprocessing is specifically: firstly, removing sharp noise of an image by using Gaussian filtering, and then removing fine interference details by using a USM sharpening enhancement algorithm, wherein the calculation method comprises the following steps:
Figure FDA0002716672710000021
where output represents an output image, orign _ image represents an original image, gaus _ image represents an image after gaussian filtering, and ω represents a USM coefficient.
5. The method for multi-target tracking based on head and torso features of claim 3, wherein the fusion method is as follows:
Figure FDA0002716672710000022
Figure FDA0002716672710000023
Figure FDA0002716672710000024
wherein c issrcRepresenting confidence, x, of key points of the body in the original imagesrcX-coordinate, y representing key points of a human body in an original imagesrcY-coordinate representing key points of the human body in the original image, where cflipRepresenting confidence, x, of key points of a human body in a flip imageflipX-coordinate, y representing key points of a human body in a flip imageflipRepresenting a flipped imageAnd the y coordinate of the middle human key point, final _ x and final _ y respectively represent the x and y coordinates of the final fused human key point, and final _ c represents the confidence coefficient of the final fused human key point.
6. The method for multi-target tracking based on head-torso features of claim 1, wherein the step S3 is specifically performed by:
step S31: screening the selected key points, and adopting a screening scheme based on confidence coefficient to enable the detection confidence coefficient threshold of the human key points to be TkpDeleting the human body key points meeting the following requirements:
joint_c＜Tkp
step S32: combining the screened key points, enabling the screened human body key point set to be Q, traversing the set Q, sequencing the whole set Q according to the sizes of x coordinates and y coordinates, finding out the human body key points at the top, the bottom, the left and the right in the set, and obtaining the minimum rectangular convex hull in the set, wherein the image content in the rectangular convex hull is the head and trunk characteristics of the target.
7. The method for multi-target tracking based on head-torso features of claim 1, wherein the step S4 is specifically performed by:
step S41: let the set of targets in the tracked state be OtrackThe set comprises all targets in a tracking state from a first frame of the video to a current frame;
step S42: traversing a set of targets O in a state being trackedtrackPerforming IOU and OKS calculation on the newly obtained head and trunk characteristics and all targets in the set to confirm whether the target is in a tracking state;
step S43: calculating the fusion metric value of the bounding box of the tracked target and the newly obtained head and trunk characteristic bounding box in the target set in the tracked state, wherein the calculation method is as follows
Figure FDA0002716672710000031
If FF is greater than 0.5, the target is considered to exist, and the target does not need to be initialized to a new target again, and the step S1 is executed; otherwise the feature is identified as belonging to a new target, proceeding to the next step;
step S44: inputting new head and trunk characteristics into a tracker for initialization, and adding a target set O in a tracked statetrackGo to step S1.
8. The method for multi-target tracking based on head torso features of claim 7, wherein the IOU and OKS calculations are performed as follows:
Figure FDA0002716672710000041
Figure FDA0002716672710000042
wherein A represents the area of the bounding box of the first object and B represents the area of the bounding box of the second object; the calculation method of the bounding box area is that the length of the rectangle is multiplied by the width of the rectangle; viszIndicates the visibility of the z-th keypoint (greater than 0 indicating visibility), dis indicates the Euclidean distance, scale, of the existing and detected human keypoints2The square root, σ, representing the size of the area occupied by these key pointszA normalization factor representing the z-th human body key point.
CN202011076008.8A 2020-10-10 2020-10-10 A method for multi-target tracking based on head and torso features Active CN112200076B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011076008.8A CN112200076B (en) 2020-10-10 2020-10-10 A method for multi-target tracking based on head and torso features

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011076008.8A CN112200076B (en) 2020-10-10 2020-10-10 A method for multi-target tracking based on head and torso features

Publications (2)

Publication Number Publication Date
CN112200076A true CN112200076A (en) 2021-01-08
CN112200076B CN112200076B (en) 2023-02-21

Family

ID=74013249

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011076008.8A Active CN112200076B (en) 2020-10-10 2020-10-10 A method for multi-target tracking based on head and torso features

Country Status (1)

Country Link
CN (1) CN112200076B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114005167A (en) * 2021-11-29 2022-02-01 西安电子科技大学 Remote sight estimation method and device based on human skeleton key points
CN117953015A (en) * 2024-03-26 2024-04-30 武汉工程大学 Multi-pedestrian tracking method, system, device and medium based on video super-resolution

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108416258A (en) * 2018-01-23 2018-08-17 华侨大学 A kind of multi-human body tracking method based on human body model
CN109522843A (en) * 2018-11-16 2019-03-26 北京市商汤科技开发有限公司 A kind of multi-object tracking method and device, equipment and storage medium
WO2019232894A1 (en) * 2018-06-05 2019-12-12 中国石油大学(华东) Complex scene-based human body key point detection system and method
CN110674785A (en) * 2019-10-08 2020-01-10 中兴飞流信息科技有限公司 Multi-person posture analysis method based on human body key point tracking
CN111062239A (en) * 2019-10-15 2020-04-24 平安科技（深圳）有限公司 Human target detection method, device, computer equipment and storage medium

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108416258A (en) * 2018-01-23 2018-08-17 华侨大学 A kind of multi-human body tracking method based on human body model
WO2019232894A1 (en) * 2018-06-05 2019-12-12 中国石油大学(华东) Complex scene-based human body key point detection system and method
CN109522843A (en) * 2018-11-16 2019-03-26 北京市商汤科技开发有限公司 A kind of multi-object tracking method and device, equipment and storage medium
CN110674785A (en) * 2019-10-08 2020-01-10 中兴飞流信息科技有限公司 Multi-person posture analysis method based on human body key point tracking
CN111062239A (en) * 2019-10-15 2020-04-24 平安科技（深圳）有限公司 Human target detection method, device, computer equipment and storage medium

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
AHMAD JALAL AT EL.: "Human Body Parts Estimation and Detection for Physical Sports Movements", 《2019 2ND INTERNATIONAL CONFERENCE ON COMMUNICATION, COMPUTING AND DIGITAL SYSTEMS (C-CODE)》 *
周亮 等: "基于跟踪—关联模块的多目标跟踪方法研究", 《西南大学学报(自然科学版)》 *
宫法明等: "基于时空双分支网络的人体动作识别研究", 《计算机技术与发展》 *
马旋等: "基于人脸检测和关键点识别的快速人体组件划分", 《计算机应用与软件》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114005167A (en) * 2021-11-29 2022-02-01 西安电子科技大学 Remote sight estimation method and device based on human skeleton key points
CN117953015A (en) * 2024-03-26 2024-04-30 武汉工程大学 Multi-pedestrian tracking method, system, device and medium based on video super-resolution

Also Published As

Publication number Publication date
CN112200076B (en) 2023-02-21

Similar Documents

Publication Publication Date Title
CN111462200B (en) A cross-video pedestrian positioning and tracking method, system and device
WO2020042419A1 (en) Gait-based identity recognition method and apparatus, and electronic device
US6826292B1 (en) Method and apparatus for tracking moving objects in a sequence of two-dimensional images using a dynamic layered representation
CN103295242B (en) A kind of method for tracking target of multiple features combining rarefaction representation
CN111639551A (en) Online multi-target tracking method and system based on twin network and long-short term clues
CN109543641B (en) Multi-target duplicate removal method for real-time video, terminal equipment and storage medium
CN111445389A (en) A fast stitching method for high-resolution images with wide viewing angle
CN110986969B (en) Map fusion method and device, equipment and storage medium
CN109766796B (en) Deep pedestrian detection method for dense crowd
CN101950426A (en) Vehicle relay tracking method in multi-camera scene
CN114782499A (en) A method and device for image static region extraction based on optical flow and view geometry constraints
CN108257155B (en) Extended target stable tracking point extraction method based on local and global coupling
CN107248174A (en) A kind of method for tracking target based on TLD algorithms
CN115131420A (en) Visual SLAM method and device based on key frame optimization
CN109146920A (en) A kind of method for tracking target that insertion type is realized
CN114926859B (en) A method for multi-target pedestrian tracking in dense scenes combined with head tracking
CN113608663A (en) Fingertip tracking method based on deep learning and K-curvature method
CN111160291A (en) Human eye detection method based on depth information and CNN
CN112200076A (en) Method for carrying out multi-target tracking based on head and trunk characteristics
CN117036404A (en) Monocular thermal imaging simultaneous positioning and mapping method and system
CN111709893A (en) An Improved ORB-SLAM2 Algorithm Based on Information Entropy and Sharpening Adjustment
CN108319961A (en) A kind of image ROI rapid detection methods based on local feature region
CN110766728B (en) Combined image feature accurate matching method based on deep learning
CN118225096A (en) Multi-sensor SLAM method based on dynamic feature point elimination and loop detection
CN114550219B (en) Pedestrian tracking method and device

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