CN108573496B - Multi-target tracking method based on LSTM network and deep reinforcement learning - Google Patents
Multi-target tracking method based on LSTM network and deep reinforcement learning Download PDFInfo
- Publication number
- CN108573496B CN108573496B CN201810271429.2A CN201810271429A CN108573496B CN 108573496 B CN108573496 B CN 108573496B CN 201810271429 A CN201810271429 A CN 201810271429A CN 108573496 B CN108573496 B CN 108573496B
- Authority
- CN
- China
- Prior art keywords
- target
- tracking
- time
- reinforcement learning
- deep reinforcement
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000002787 reinforcement Effects 0.000 title claims abstract description 26
- 238000001514 detection method Methods 0.000 claims abstract description 29
- 239000013598 vector Substances 0.000 claims abstract description 29
- 238000005516 engineering process Methods 0.000 claims abstract description 14
- 230000008859 change Effects 0.000 claims abstract description 11
- 230000008569 process Effects 0.000 claims abstract description 11
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims abstract description 7
- 230000009471 action Effects 0.000 claims description 22
- 239000003795 chemical substances by application Substances 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000007704 transition Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- RTAQQCXQSZGOHL-UHFFFAOYSA-N Titanium Chemical compound [Ti] RTAQQCXQSZGOHL-UHFFFAOYSA-N 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a multi-target tracking method based on an LSTM network and deep reinforcement learning, which adopts a target detector to detect each frame of image in a video to be detected and outputs a detection result
Constructing a plurality of single target trackers based on the deep reinforcement learning technology, wherein each single target tracker comprises a convolutional neural network and a full connection layer, and the convolutional neural network is constructed on the basis of a VGG-16 network and outputs the tracking result of each single target tracker
Computing similarity matrices for data correlations
Constructing a data association module based on an LSTM network, and inputting a similarity matrix to obtain a distribution probability vector
And the matching probability between the ith target and the detection result j is obtained, and the target detection result with the maximum matching probability is used as the tracking result of the ith target. The method and the device are not influenced by mutual shielding, similar appearance and continuous change of quantity in the multi-target tracking process, and improve the multi-target tracking accuracy and the multi-target tracking precision.
Description
Technical Field
The invention belongs to the field of computer vision, relates to a video multi-target tracking method, and particularly relates to a multi-target tracking method based on an LSTM network and deep reinforcement learning.
Background
Multi-target tracking is a hot problem in the field of computer vision, and plays an important role in many application fields, such as: artificial intelligence, virtual reality, unmanned, etc. Although a great deal of related work exists in the early stage, the multi-target tracking is still a challenging problem due to the problems of frequent shielding, similar appearance of multiple targets, continuous change of the number of targets and the like in the multi-target tracking process.
In recent years, detection-based multi-target tracking methods have been successful, and divide multi-target tracking into two parts, namely multi-target detection and data association. Deep learning and deep reinforcement learning technologies have recently been widely applied in the computer field, but no relevant research results exist in the multi-target tracking technology.
Disclosure of Invention
The purpose of the invention is as follows: in order to overcome the technical defects that a manually designed model is not comprehensive enough and the tracking result is not accurate enough in the prior art, the invention provides a multi-target tracking method based on an LSTM network and deep reinforcement learning.
The technical scheme is as follows: a multi-target tracking method based on an LSTM network and deep reinforcement learning comprises the following steps:
(1) detecting each frame of image in the video to be detected by adopting a target detector, outputting a detection result, and setting the detection result of the image corresponding to the time t as a set
The jth detection result of the corresponding image at the time t, and N is the total number of detections;
(2) constructing a plurality of single target trackers based on the deep reinforcement learning technology, wherein each single target tracker comprises a convolutional neural network and a full connection layer, the convolutional neural network is constructed on the basis of a VGG-16 network, the tracking result of each single target tracker is output, and the tracking result of the image corresponding to the time t is set as a set
The output result of the ith single-target tracker of the corresponding image at the time t is obtained, and M is the total number of targets which can be tracked at the time t simultaneously;
(3) according to step (1)
And step (2)
Computing similarity matrices for data correlations
Is that
And
the euclidean distance between them,
(4) constructing a data association module based on an LSTM network and inputting a similarity matrix
Obtaining an assigned probability vector
For the match probability between the ith target and all possible target detections,
and the matching probability between the ith target and the detection result j is obtained, and the target detection result with the maximum matching probability is used as the tracking result of the ith target.
Further, the target detector in the step (1) adopts YOLO V2.
Further, the detection result output by the target detector in the step (1) and the output result of the target tracker in the step (2) are both four-dimensional vectors,
wherein (x)t',yt') center coordinates of the target tracking rectangle in the target detector, wt' Width of rectangular frame for target tracking in target detector, ht' tracking the height of a rectangular box for an object in the object detector; (x)t,yt) Tracking center coordinates, w, of a rectangular frame for a target in a single target trackertWidth, h, of a rectangular frame for tracking a target in a single target trackertThe height of the rectangular box is tracked for the target in the single target tracker.
Further, the specific method for outputting the tracking result by the single-target tracker in the step (2) is as follows:
the single target tracker based on the deep reinforcement learning technology takes each target as an intelligent agent, trains the intelligent agent by utilizing the deep reinforcement learning technology, and determines actions according to feedback given by the intelligent agent according to the self state and the environment;
action a at time ttThe vector is a seven-dimensional vector which comprises motion in two horizontal directions, motion in two vertical directions, size change and search termination action, and the state vector at the moment t is set as st＝(pt,vt)，vtIs a vector of historical actions, state vector s at time t +1t+1＝(pt+1,vt+1) From the state s at time tt＝(pt,vt) Determining, based on the state transition equation, that p ist+1＝ft(pt,at) And the equation of motion change vt+1＝fv(vt,at) Predicting a state vector at the t +1 moment according to the motion at the t moment and the state vector at the t moment;
in the training process, the intelligent body receives a feedback signal rtAt each iteration instant in the tracking process, rt0, when the termination of the search action is selected at the termination time T, the feedback signal rTIs a threshold function of the intersection ratio IoU:
wherein IoU (p)T,g)＝area(pT∩g)/area(pT∪ g), g is the true value of the image block, i.e. the true position of the target calibrated manually, and tau is the threshold set manually.
Further, the LSTM network in step (4) includes the input parameters: step i hidden state hiStep i cell status ciSimilarity matrix
Hidden state h of output parameter step i +1i+1Step i +1 cell status ci+1Assigning probability vectors
First to the hidden state hiCell state ciInitializing, gradually inputting the hidden state h of the ith stepiStep i cell status ciAnd similar matrices
Outputting the hidden state h of the step i +1i+1Step i +1 cell status ci+1And
has the advantages that: the LSTM network and depth reinforcement learning technology is applied to the video multi-target tracking method for the first time, the technical defects that a manually designed model is not comprehensive enough and the tracking result is not accurate enough are overcome, the influence of mutual shielding, similar appearance and continuous change of quantity in the multi-target tracking process is avoided, and the multi-target tracking accuracy are improved.
Drawings
FIG. 1 is a system diagram of the multi-target tracking method based on the LSTM network and the deep reinforcement learning according to the present invention;
FIG. 2 is a schematic diagram of a single target tracker;
FIG. 3 is a block diagram of deep reinforcement learning;
fig. 4 is a schematic diagram of an LSTM network.
Detailed Description
The invention is further described with reference to the following figures and specific examples.
As shown in fig. 1, the multi-target tracking method based on LSTM network and deep reinforcement learning includes the following steps:
(1) detecting each frame of image in the video to be detected by adopting a YOLO V2 target detector, outputting a detection result, and setting the detection result of the image corresponding to the time t as a set
The jth detection result of the corresponding image at the time t, and N is the total number of detections;
(2) as shown in FIG. 2, a plurality of single target trackers based on the deep reinforcement learning technology are constructed, each single target tracker comprises a convolutional neural network CNN and a full connection layer FC, the convolutional neural network is constructed on the basis of a VGG-16 network, the VGG-16 belongs to the prior art, and the deep reinforcement learning technology is widely applied to the deep learning method. The CNN network designed by the invention comprises 5 pooling layers, namely Conv1-2, Conv2-2, Conv3-3, Conv4-3 and Conv5-3, and characteristics output by Conv3-3, Conv4-3 and Conv5-3 are used as target expression characteristics in a tracking process. Outputting traces for each single target trackerAs a result, let the tracking result of the corresponding image at time t be the set
The output result of the ith single-target tracker of the corresponding image at the time t is obtained, and M is the total number of targets which can be tracked at the time t simultaneously;
as shown in fig. 3, the single target tracker based on the deep reinforcement learning technology regards each target as an agent, and trains the agent by using the deep reinforcement learning technology, and each agent determines an action according to the feedback given by the agent's own state and environment;
the action set (action) A adopted by us is composed of 6 actions in different directions and 1 action for terminating search, including horizontal two-direction movement { right, left }, vertical two-direction movement { up, down }, size change { scaleup, scale down }, and action for terminating search, that is, action a at time ttConsisting of a 7-dimensional vector. Let the state vector at time t be st＝(pt,vt)，ptRepresentative image blocks, vtIs a vector of historical actions, this patent stores 10 historical actions, which means vtIs a 70-dimensional vector. State vector s at time t +1t+1＝(pt+1,vt+1) From the state s at time tt＝(pt,vt) Determining, based on the state transition equation, that p ist+1＝ft(pt,at) And the equation of motion change vt+1＝fv(vt,at) Predicting the state vector at the t +1 moment according to the motion at the t moment and the state vector at the t moment, as shown in FIG. 2; the specific prediction method is described as follows: let ptIs [ x ]t,yt,wt,ht]The state transition equation includes: Δ xt＝αwt,Δyt＝αhtα is 0.03, p is calculated by using the action change equationt+1If act atBy "left", i.e. pt+1Is [ x ]t-Δxt,yt,wt,ht](ii) a If action atTo "scaleup", then pt+1Is [ x ]t,yt,wt+Δxt,ht+Δyt]。
In the training process, the intelligent body receives a feedback signal rtFeedback signal rtThe effect of (a) is to tell the agent how to move or whether to terminate the action. At each iteration instant in the tracking process, rt0, when the termination of the search action is selected at the termination time T, the feedback signal rTIs a threshold function of IoU (Intersection-over-Union):
wherein IoU (p)T,g)＝area(pT∩g)/area(pT∪ g), g is the real value of the image block (the group path), i.e. the real position of the target calibrated manually, and tau is the threshold set manually.
(3) The detection result output by the target detector in the step (1) and the output result of the target tracker in the step (2) are both four-dimensional vectors,
wherein (x)t',yt') center coordinates of the target tracking rectangle in the target detector, wt' Width of rectangular frame for target tracking in target detector, ht' tracking the height of a rectangular box for an object in the object detector; (x)t,yt) Tracking center coordinates, w, of a rectangular frame for a target in a single target trackertWidth, h, of a rectangular frame for tracking a target in a single target trackertThe height of the rectangular box is tracked for the target in the single target tracker. According to step (1)
And step (2)
Computing similarity matrices for data correlations
Is that
And
the euclidean distance between them,
data-associative similarity matrix for measuring output of single-target tracker
And the output of the target detector
The degree of similarity between them;
(4) the data association module is constructed based on an LSTM network, and the LSTM network comprises the following input parameters: step i hidden state hiStep i cell status ciSimilarity matrix
Hidden state h of output parameter step i +1i+1Step i +1 cell status ci+1Assigning probability vectors
First to the hidden state hiCell state ciInitializing, gradually inputting the hidden state h of the ith stepiStep i cell status ciAnd similar matrices
Output the (i + 1) th step of hiddenHidden state hi+1Step i +1 cell status ci+1And
for the match probability between the ith target and all possible target detections,
and the matching probability between the ith target and the detection result j is obtained, and the target detection result with the maximum matching probability is used as the tracking result of the ith target.
In order to verify the technical effect of the method, the following experiments were carried out:
the Windows 10 operating system was used for experiments, MATLAB R2016a was used as the software platform, and the computer was configured primarily as Intel (R) core (TM) i7-4712MQ CPU @3.40GHz (with 16G memory) with TITAN GPU (12.00GB memory).
In order to measure the performance of the tracking method, 4 indexes of multi-target tracking accuracy (MOTA), multi-target tracking accuracy (MOTP), false alarm rate (FP) and identification switching (IDSW) are selected for comparison, and a table 1 lists comparison results between the method (LSTM _ DRL) and other 4 methods in the prior art, and data are based on 11 test videos.
TABLE 1 comparative results.
Experiments show that the method can overcome the defect that the tracking result in the prior art is not accurate enough, is not influenced by mutual shielding, similar appearance and continuous change of quantity in the multi-target tracking process, and can improve the multi-target tracking accuracy and the multi-target tracking precision.
Claims (4)
1. A multi-target tracking method based on an LSTM network and deep reinforcement learning is characterized by comprising the following steps:
(1) detecting each frame of image in the video to be detected by adopting a target detector, outputting a detection result, and setting the detection result of the image corresponding to the time t as a set
The jth detection result of the corresponding image at the time t, and N is the total number of detections;
(2) constructing a plurality of single target trackers based on the deep reinforcement learning technology, wherein each single target tracker comprises a convolutional neural network and a full connection layer, the convolutional neural network is constructed on the basis of a VGG-16 network, the tracking result of each single target tracker is output, and the tracking result of the image corresponding to the time t is set as a set
The output result of the ith single-target tracker of the corresponding image at the time t is obtained, and M is the total number of targets which can be tracked at the time t simultaneously;
the specific method for outputting the tracking result by the single-target tracker in the step (2) is as follows:
the single target tracker based on the deep reinforcement learning technology takes each target as an intelligent agent, trains the intelligent agent by utilizing the deep reinforcement learning technology, and determines actions according to feedback given by the intelligent agent according to the self state and the environment;
action a at time ttThe vector is a seven-dimensional vector which comprises motion in two horizontal directions, motion in two vertical directions, size change and search termination action, and the state vector at the moment t is set as st＝(pt,vt)，vtIs the direction of historical actionsAmount, ptFor the tracking target image block at time t, the state vector s at time t +1t+1＝(pt+1,vt+1) From the state s at time tt＝(pt,vt) Determination of pt+1The image block of the tracked target at the time of t +1 is p according to a state conversion equationt+1＝ft(pt,at) And the equation of motion change vt+1＝fv(vt,at) Predicting a state vector at the t +1 moment according to the motion at the t moment and the state vector at the t moment;
in the training process, the intelligent body receives a feedback signal rtAt each iteration instant in the tracking process, rt0, when the termination of the search action is selected at the termination time T, the feedback signal rTIs a threshold function of the intersection ratio IoU:
wherein IoU (p)T,g)＝area(pT∩g)/area(pT∪ g), g is the real value of the image block, i.e. the real position of the target calibrated manually, tau is the threshold set manually, pTA tracking target image block at the termination time T;
(3) according to step (1)
And step (2)
Computing similarity matrices for data correlations
Is that
And
the euclidean distance between them,
(4) constructing a data association module based on an LSTM network and inputting a similarity matrix
Obtaining an assigned probability vector
For the match probability between the ith target and all possible target detections,
and the matching probability between the ith target and the detection result j is obtained, and the target detection result with the maximum matching probability is used as the tracking result of the ith target.
2. The LSTM network and deep reinforcement learning-based multi-target tracking method of claim 1, wherein the target detector in step (1) adopts YOLO V2.
3. The LSTM network and deep reinforcement learning based multi-target tracking method of claim 1, wherein the detection result outputted by the target detector in step (1) and the output result of the target tracker in step (2) are four-dimensional vectors,
wherein, (x't,y't) Tracking center coordinates, w ', of rectangular box for targets in target detector'tTracking width, h 'of rectangular box for target in target detector'tTracking the height of the rectangular box for the target in the target detector; (x)t,yt) Tracking center coordinates, w, of a rectangular frame for a target in a single target trackertWidth, h, of a rectangular frame for tracking a target in a single target trackertThe height of the rectangular box is tracked for the target in the single target tracker.
4. The LSTM network and deep reinforcement learning-based multi-target tracking method according to claim 1, wherein the LSTM network in step (4) includes input parameters: step i hidden state hiStep i cell status ciSimilarity matrix
Hidden state h of output parameter step i +1i+1Step i +1 cell status ci+1Assigning probability vectors
First to the hidden state hiCell state ciInitializing, gradually inputting the hidden state h of the ith stepiStep i cell status ciAnd similar matrices
Outputting the hidden state h of the step i +1i+1Step i +1 cell status ci+1And
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810271429.2A CN108573496B (en) | 2018-03-29 | 2018-03-29 | Multi-target tracking method based on LSTM network and deep reinforcement learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810271429.2A CN108573496B (en) | 2018-03-29 | 2018-03-29 | Multi-target tracking method based on LSTM network and deep reinforcement learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108573496A CN108573496A (en) | 2018-09-25 |
| CN108573496B true CN108573496B (en) | 2020-08-11 |
Family
ID=63574660
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810271429.2A Active CN108573496B (en) | 2018-03-29 | 2018-03-29 | Multi-target tracking method based on LSTM network and deep reinforcement learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108573496B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109360226B (en) * | 2018-10-17 | 2021-09-24 | 武汉大学 | Multi-target tracking method based on time series multi-feature fusion |
| CN109685100B (en) * | 2018-11-12 | 2024-05-10 | 平安科技（深圳）有限公司 | Character recognition method, server and computer readable storage medium |
| CN109682392B (en) * | 2018-12-28 | 2020-09-01 | 山东大学 | Visual navigation method and system based on deep reinforcement learning |
| CN109816701B (en) * | 2019-01-17 | 2021-07-27 | 北京市商汤科技开发有限公司 | Target tracking method and device and storage medium |
| CN110544268B (en) * | 2019-07-29 | 2023-03-24 | 燕山大学 | Multi-target tracking method based on structured light and SiamMask network |
| CN111127513B (en) * | 2019-12-02 | 2024-03-15 | 北京交通大学 | Multi-target tracking method |
| CN111027505B (en) * | 2019-12-19 | 2022-12-23 | 吉林大学 | Hierarchical multi-target tracking method based on significance detection |
| CN111354023A (en) * | 2020-03-09 | 2020-06-30 | 中振同辂(江苏)机器人有限公司 | Camera-based visual multi-target tracking method |
| CN112053385B (en) * | 2020-08-28 | 2023-06-02 | 西安电子科技大学 | Remote sensing video shielding target tracking method based on deep reinforcement learning |
| CN112381021B (en) * | 2020-11-20 | 2022-07-12 | 安徽一视科技有限公司 | Personnel detection counting method based on deep learning |
| CN113762231B (en) * | 2021-11-10 | 2022-03-22 | 中电科新型智慧城市研究院有限公司 | End-to-end multi-pedestrian posture tracking method and device and electronic equipment |
| CN117788511B (en) * | 2023-12-26 | 2024-06-25 | 兰州理工大学 | Multi-expansion target tracking method based on deep neural network |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022239A (en) * | 2016-05-13 | 2016-10-12 | 电子科技大学 | Multi-target tracking method based on recurrent neural network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10242266B2 (en) * | 2016-03-02 | 2019-03-26 | Mitsubishi Electric Research Laboratories, Inc. | Method and system for detecting actions in videos |
-
2018
- 2018-03-29 CN CN201810271429.2A patent/CN108573496B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022239A (en) * | 2016-05-13 | 2016-10-12 | 电子科技大学 | Multi-target tracking method based on recurrent neural network |
Non-Patent Citations (2)
| Title |
|---|
| Deep Reinforcement Learning for Visual Object Tracking in Videos;Da Zhang等;《https://arxiv.org/pdf/1701.08936v2.pdf》;20170413;第1-5页 * |
| Online Multi-Target Tracking Using Recurrent Neural Networks;Anton Milan等;《https://arxiv.org/pdf/1604.03635.pdf》;20161210;第1-7页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108573496A (en) | 2018-09-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108573496B (en) | Multi-target tracking method based on LSTM network and deep reinforcement learning | |
| Chiu et al. | Probabilistic 3D multi-modal, multi-object tracking for autonomous driving | |
| CN111127513B (en) | Multi-target tracking method | |
| CN109800689B (en) | Target tracking method based on space-time feature fusion learning | |
| WO2020215492A1 (en) | Multi-bernoulli multi-target video detection and tracking method employing yolov3 | |
| CN107516321B (en) | Video multi-target tracking method and device | |
| CN111932580A (en) | Road 3D vehicle tracking method and system based on Kalman filtering and Hungary algorithm | |
| CN112052802B (en) | Machine vision-based front vehicle behavior recognition method | |
| CN112836640A (en) | Single-camera multi-target pedestrian tracking method | |
| CN104200488A (en) | Multi-target tracking method based on graph representation and matching | |
| CN104574439A (en) | Kalman filtering and TLD (tracking-learning-detection) algorithm integrated target tracking method | |
| CN109191488B (en) | Target tracking system and method based on CSK and TLD fusion algorithm | |
| CN110363165B (en) | Multi-target tracking method and device based on TSK fuzzy system and storage medium | |
| CN110349188B (en) | Multi-target tracking method, device and storage medium based on TSK fuzzy model | |
| CN111626194A (en) | Pedestrian multi-target tracking method using depth correlation measurement | |
| CN108022254A (en) | A kind of space-time contextual target tracking based on sign point auxiliary | |
| CN108898612A (en) | Multi-object tracking method based on the enhancing study of multiple agent depth | |
| CN112036381B (en) | Visual tracking method, video monitoring method and terminal equipment | |
| WO2018227491A1 (en) | Method and device for association of fuzzy data of multiple targets in video | |
| CN111639570B (en) | Online multi-target tracking method based on motion model and single-target clue | |
| CN116977367A (en) | Campus multi-target tracking method based on transform and Kalman filtering | |
| CN113971688B (en) | Anchor-free multi-target tracking method for enhancing ID re-identification | |
| Wang et al. | Effective multiple pedestrian tracking system in video surveillance with monocular stationary camera | |
| Qing et al. | A novel particle filter implementation for a multiple-vehicle detection and tracking system using tail light segmentation | |
| CN111429481A (en) | Target tracking method, device and terminal based on adaptive expression |
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 | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20180925 Assignee: Huaian xiaobaihu coating Engineering Co.,Ltd. Assignor: HUAIYIN INSTITUTE OF TECHNOLOGY Contract record no.: X2020980006698 Denomination of invention: Multi target tracking method based on LSTM network and deep reinforcement learning Granted publication date: 20200811 License type: Common License Record date: 20201010 |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20231129 Address after: 200000 Building 1, 377 Nanqiao Road, Nanqiao Town, Fengxian District, Shanghai Patentee after: Shanghai Xinghao Information Technology Co.,Ltd. Address before: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province Patentee before: Dragon totem Technology (Hefei) Co.,Ltd. Effective date of registration: 20231129 Address after: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province Patentee after: Dragon totem Technology (Hefei) Co.,Ltd. Address before: 223005 Jiangsu Huaian economic and Technological Development Zone, 1 East Road. Patentee before: HUAIYIN INSTITUTE OF TECHNOLOGY |