CN113205108A - YOLOv 4-based multi-target vehicle detection and tracking method - Google Patents
YOLOv 4-based multi-target vehicle detection and tracking method Download PDFInfo
- Publication number
- CN113205108A CN113205108A CN202011206816.1A CN202011206816A CN113205108A CN 113205108 A CN113205108 A CN 113205108A CN 202011206816 A CN202011206816 A CN 202011206816A CN 113205108 A CN113205108 A CN 113205108A
- Authority
- CN
- China
- Prior art keywords
- boxes
- detection
- target
- yolov4
- similarity
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a multi-target vehicle detection tracking method based on YOLOv4, which comprises the steps of firstly, optimizing the prediction of an anchor box through a k-means clustering algorithm so that YOLOv4 can better adapt to the requirements of a vehicle data set; secondly, the target detection network of YOLOv4 is improved, and the detection precision is effectively improved; and finally, solving the data association problem between the prediction result and the tracking result by adopting Kalman filtering and Hungarian algorithm, and effectively reducing the ID Switch phenomenon by combining target motion information and apparent information as the total association cost. The method not only improves the accuracy of detecting multiple targets and weak and small targets in a complex scene, but also improves the robustness and adaptability of a target tracking algorithm.
Description
Technical Field
The invention relates to the technical field of target tracking, in particular to a multi-target vehicle detection and tracking method based on YOLOv 4.
Background
Target detection and tracking are hot problems in the field of computer vision, and have important significance in the aspects of intelligent video monitoring, intelligent traffic, robot vision navigation, military guidance and the like. In recent years, with the continuous development of deep learning, the convolutional neural network is widely applied to target detection and tracking, a tracking algorithm which adopts a deep learning network is derived, and great success is achieved in the field of target detection and tracking.
Currently, the target detection algorithm is generally divided into four steps: firstly, target detection, namely selecting a target boundary frame by using a target detection network; secondly, extracting features, namely extracting apparent information and motion information of the target by establishing an apparent feature extraction network, and then predicting the position of the next frame of the target; thirdly, calculating the similarity through the incidence matrix, calculating the incidence matrix by using the apparent characteristics and the position characteristics, and then calculating the similarity of the two frames of targets before and after; and fourthly, matching the targets, associating the target detected by the current frame with the tracked target, and distributing the same ID for the tracked target after the association is successful.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a YOLOv 4-based multi-target vehicle detection and tracking method, which solves the problem of false detection and missed detection of multiple targets and weak targets in target tracking and improves the robustness and adaptability of a target tracking algorithm.
In order to solve the technical problems, the invention provides the following technical scheme:
a multi-target vehicle detection and tracking method based on YOLOv4 comprises the following steps:
s1, generating an anchor box through a k-means clustering algorithm, wherein the anchor box is widely used for setting the initial size of the bounding box in a single-layer target detection algorithm, and is superior to other unsupervised learning algorithms;
s2, improving a target detection network of YOLOv4, continuously increasing the scale on the basis of the fusion of the three scale features of the original YOLOv4, and adding the feature maps into four different scales;
s3, carrying out vehicle Detection on the video frames through an improved YOLOv4 target Detection network to obtain all detected target vehicle frames Detection boxes;
s4, predicting the state of the vehicle in the target vehicle detection frame through a Kalman filter to obtain corresponding target tracking frames;
s5, constructing a cost matrix between the Detection boxes and the Track boxes by utilizing the motion similarity and the apparent similarity between all the Detection boxes and the Track boxes;
and S6, performing relevance matching on the relevance cost in the relevance cost matrix according to the Hungarian algorithm, calculating the matching degree between the two frames before and after, further determining a tracking result, allocating the ID of a target to each object, and realizing multi-target vehicle detection.
Further, the step S1 of generating an anchor box by a k-means clustering algorithm specifically includes the following steps:
s1.1, acquiring a real boundary box of a target on a data set;
s1.2, a k-means algorithm randomly selects k bounding boxes as clustering heads to initialize a normalization process, redistributes clusters around the nearest centroid, and updates according to a certain threshold value until k anchor boxes are generated after convergence.
Further, the step S2 improves the target detection network of YOLOv4, and specifically includes the following steps:
s2.1, modifying a backbone network CSPDarknet-53 of YOLOv4, and adding a feature layer to ensure that the CSPDarknet-53 has four feature layers;
s2.2, inputting the last layer of feature layer into an SPP structure to carry out four times of maximum pooling operation, wherein the sizes of the pooled nuclei of the maximum pooling are respectively 13x13, 9x9, 5x5 and 1x 1;
and S2.3, inputting the four feature layers into the PANET structure to realize top-to-bottom feature extraction and bottom-to-top feature extraction.
And S2.4, finally, predicting the obtained features by using Yoloidea.
Further, the step S5 specifically includes the following steps of constructing a cost matrix between the Detection boxes and the Track boxes by using motion similarities and apparent similarities between all the Detection boxes and the Track boxes:
s5.1, measuring the distance between the Track boxes and the Detection boxes by using the squared Mahalanobis distance to calculate the similarity between the Track boxes and the Detection boxes, wherein the specific formula is as follows:
djrepresents the jth Track boxes, yjRepresents the ith Track boxes,represents the covariance of d and y;
equation (2) is an indicator that compares the Mahalanobis distance to a threshold of chi-squared distribution, t(1)9.4877, measuring the matching degree of Detection boxes and Track boxes by a threshold value;
s5.2, measuring the distance between the apparent features by using the cosine distance, wherein the calculation formula is as follows:
the cosine similarity is calculated, the cosine distance is 1-cosine similarity, the apparent characteristics of Track boxes and the apparent characteristics corresponding to Detection boxes are measured through the cosine distance, and the formula (4) is also an indicator;
s5.3, obtaining a correlation cost matrix by weighting the similarity and the apparent similarity of the moving target, wherein the formula is as follows:
ci,j=λd(1)(i,j)+(1-λ)d(2)(i,j) (5)
where λ is a hyper-parameter and defaults to 0.
Further, the step S6 is to perform correlation matching on the correlation cost in the correlation cost matrix according to the hungarian algorithm, calculate the matching degree between the two frames before and after, further determine the tracking result, assign an ID of a target to each object, and implement multi-target vehicle detection specifically includes the following steps:
s6.1, setting a similarity threshold value, and comparing the similarity threshold value with the cost matrix calculated in the step S5;
and S6.2, allocating the same ID to the targets in the Detection boxes and the Track boxes corresponding to the cost matrix with the similarity threshold value, and taking the same ID as a group of tracking results.
The invention has the following beneficial effects:
(1) the prediction of the anchor box is optimized through a k-means clustering algorithm, so that YOLOv4 is more suitable for the requirement of a vehicle data set, and the precision of Detection boxes is improved;
(2) the CSPDarknet-53 of the backbone network of YOLOv4 is modified, and a feature layer is added, so that the CSPDarknet-53 has four feature layers, and the detection precision of small targets is improved;
(3) the data association problem between the predicted and the tracked results is solved by using a Kalman filtering algorithm and a Hungarian algorithm, a cost matrix is generated by using the motion similarity and the apparent similarity of the target, and the ID Switch phenomenon is effectively reduced.
Drawings
FIG. 1 is a flow chart of a tracking model of the present invention;
FIG. 2 is a diagram of an improved object detection network architecture of the present invention;
FIG. 3 is a flow chart of Detection boxes and Track boxes matching for Kalman filtering and Hungarian algorithms.
Detailed Description
The following description of the embodiments of the present invention is provided in conjunction with the accompanying drawings to facilitate those skilled in the art to understand the present invention, and it is to be understood that the embodiments described herein are further intended to illustrate and not to limit the present invention, and that various changes will be apparent to those skilled in the art as long as they are within the scope and range of the present invention as defined and de-identified in the appended claims, and all inventions utilizing the concepts of the present invention are protected.
Referring to fig. 1 to 3, a method for detecting and tracking multiple targets of vehicles based on YOLOv4 includes the following steps:
s1, generating an anchor box through a k-means clustering algorithm, wherein the anchor box is widely used for setting the initial size of the bounding box in a single-layer target detection algorithm, and is superior to other unsupervised learning algorithms;
s2, improving a target detection network of YOLOv4, continuously increasing the scale on the basis of the fusion of the three scale features of the original YOLOv4, and adding the feature maps into four different scales;
s3, carrying out vehicle Detection on the video frames through an improved YOLOv4 target Detection network to obtain all detected target vehicle frames Detection boxes;
s4, predicting the state of the vehicle in the target vehicle detection frame through a Kalman filter to obtain corresponding target tracking frames;
s5, constructing a cost matrix between the Detection boxes and the Track boxes by utilizing the motion similarity and the apparent similarity between all the Detection boxes and the Track boxes;
and S6, performing relevance matching on the relevance cost in the relevance cost matrix according to the Hungarian algorithm, calculating the matching degree between the two frames before and after, further determining a tracking result, allocating the ID of a target to each object, and realizing multi-target vehicle detection.
The step S1 of generating the anchor box through the k-means clustering algorithm specifically comprises the following steps:
s1.1, acquiring a real boundary box of a target on a data set;
s1.2, a k-means algorithm randomly selects k bounding boxes as clustering heads to initialize a normalization process, redistributes clusters around the nearest centroid, and updates according to a certain threshold value until k anchor boxes are generated after convergence.
The step S2 improves the target detection network of YOLOv4, and specifically includes the following steps:
s2.1, modifying a backbone network CSPDarknet-53 of YOLOv4, and adding a feature layer to ensure that the CSPDarknet-53 has four feature layers;
s2.2, inputting the last layer of feature layer into an SPP structure to carry out four times of maximum pooling operation, wherein the sizes of the pooled nuclei of the maximum pooling are respectively 13x13, 9x9, 5x5 and 1x 1;
and S2.3, inputting the four feature layers into the PANET structure to realize top-to-bottom feature extraction and bottom-to-top feature extraction.
And S2.4, finally, predicting the obtained features by using Yoloidea.
The step S5, which utilizes motion similarities and apparent similarities between all Detection boxes and Track boxes, specifically includes the following steps:
s5.1, measuring the distance between the Track boxes and the Detection boxes by using the squared Mahalanobis distance to calculate the similarity between the Track boxes and the Detection boxes, wherein the specific formula is as follows:
djrepresents the jth Track boxes, yjRepresents the ith Track boxes,represents the covariance of d and y;
equation (2) is an indicator that compares the Mahalanobis distance to a threshold of chi-squared distribution, t(1)9.4877, measuring the matching degree of Detection boxes and Track boxes by a threshold value;
s5.2, measuring the distance between the apparent features by using the cosine distance, wherein the calculation formula is as follows:
the cosine similarity is calculated, the cosine distance is 1-cosine similarity, the apparent characteristics of Track boxes and the apparent characteristics corresponding to Detection boxes are measured through the cosine distance, and the formula (4) is also an indicator;
s5.3, obtaining a correlation cost matrix by weighting the similarity and the apparent similarity of the moving target, wherein the formula is as follows:
ci,j=λd(1)(i,j)+(1-λ)d(2)(i,j) (5)
where λ is a hyper-parameter and defaults to 0.
The step S6 described above is to perform correlation matching on the correlation cost in the correlation cost matrix according to the hungarian algorithm, calculate the matching degree between the two frames before and after, further determine the tracking result, assign an ID of a target to each object, and implement multi-target vehicle detection specifically includes the following steps:
s6.1, setting a similarity threshold value, and comparing the similarity threshold value with the cost matrix calculated in the step S5;
and S6.2, allocating the same ID to the targets in the Detection boxes and the Track boxes corresponding to the cost matrix with the similarity threshold value, and taking the same ID as a group of tracking results.
The foregoing illustrates a method for detecting and tracking multiple targets of vehicles based on YOLOv4, and the collective flowchart is shown in fig. 1.
The invention has the beneficial effects that:
(1) the prediction of the anchor box is optimized through a k-means clustering algorithm, so that YOLOv4 is more suitable for the requirement of a vehicle data set, and the precision of Detection boxes is improved;
(2) the CSPDarknet-53 of the backbone network of YOLOv4 is modified, and a feature layer is added, so that the CSPDarknet-53 has four feature layers, and the detection precision of small targets is improved;
(3) the data association problem between the predicted and the tracked results is solved by using a Kalman filtering algorithm and a Hungarian algorithm, a cost matrix is generated by using the motion similarity and the apparent similarity of the target, and the ID Switch phenomenon is effectively reduced.
Claims (4)
1. A multi-target vehicle detection and tracking method based on YOLOv4 is characterized by comprising the following steps:
s1, generating an anchor box through a k-means clustering algorithm, wherein the anchor box is widely used for setting the initial size of the bounding box in a single-layer target detection algorithm, and is superior to other unsupervised learning algorithms;
s2, improving a target detection network of YOLOv4, continuously increasing the scale on the basis of the fusion of the three scale features of the original YOLOv4, and adding the feature maps into four different scales;
s3, carrying out vehicle Detection on the video frames through an improved YOLOv4 target Detection network to obtain all detected target vehicle frames Detection boxes;
s4, predicting the state of the vehicle in the target vehicle detection frame through a Kalman filter to obtain corresponding target tracking frames;
s5, constructing a cost matrix between the Detection boxes and the Track boxes by utilizing the motion similarity and the apparent similarity between all the Detection boxes and the Track boxes;
and S6, performing relevance matching on the relevance cost in the relevance cost matrix according to the Hungarian algorithm, calculating the matching degree between the two frames before and after, further determining a tracking result, allocating the ID of a target to each object, and realizing multi-target vehicle detection.
2. The method for detecting and tracking the multiple targets based on the YOLOv4 of claim 1, wherein the step S1 of generating an anchor box through a k-means clustering algorithm specifically comprises the following steps:
s1.1, acquiring a real boundary box of a target on a data set;
s1.2, a k-means algorithm randomly selects k bounding boxes as clustering heads to initialize a normalization process, redistributes clusters around the nearest centroid, and updates according to a certain threshold value until k anchor boxes are generated after convergence.
3. The method for detecting and tracking multiple targets of vehicles based on YOLOv4 of claim 1, wherein the step S2 improves the target detection network of YOLOv4, and comprises the following steps:
s2.1, modifying a backbone network CSPDarknet-53 of YOLOv4, and adding a feature layer to ensure that the CSPDarknet-53 has four feature layers;
s2.2, inputting the last layer of feature layer into an SPP structure to carry out four times of maximum pooling operation, wherein the sizes of the pooled nuclei of the maximum pooling are respectively 13x13, 9x9, 5x5 and 1x 1;
s2.3, inputting the four feature layers into the PANET structure to realize feature extraction from top to bottom and feature extraction from bottom to top;
and S2.4, finally, predicting the obtained features by using Yoloidea.
4. The YOLOv 4-based multi-target vehicle Detection and tracking method according to claim 1, wherein the step S5 is to construct the cost matrix between the Detection boxes and the Track boxes by using the motion similarity and the apparent similarity between all the Detection boxes and the Track boxes, and specifically comprises the following steps:
s5.1, measuring the distance between the Track boxes and the Detection boxes by using the squared Mahalanobis distance to calculate the similarity between the Track boxes and the Detection boxes, wherein the specific formula is as follows:
represents the firstjA plurality of Track boxes are arranged in the base,represents the firstiA plurality of Track boxes are arranged in the base,representsdAndythe covariance of (a);
equation (2) is an indicator that compares mahalanobis distance to a threshold value for chi-squared distribution,measuring the matching degree of Detection boxes and Track boxes through a threshold value;
s5.2, measuring the distance between the apparent features by using the cosine distance, wherein the calculation formula is as follows:
the cosine similarity is calculated, the cosine distance = 1-cosine similarity, the apparent features of the Track boxes and the apparent features corresponding to the Detection boxes are measured through the cosine distance, and the formula (4) is also an indicator;
s5.3, obtaining a correlation cost matrix by weighting the similarity and the apparent similarity of the moving target, wherein the formula is as follows:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011206816.1A CN113205108A (en) | 2020-11-02 | 2020-11-02 | YOLOv 4-based multi-target vehicle detection and tracking method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011206816.1A CN113205108A (en) | 2020-11-02 | 2020-11-02 | YOLOv 4-based multi-target vehicle detection and tracking method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113205108A true CN113205108A (en) | 2021-08-03 |
Family
ID=77025068
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011206816.1A Pending CN113205108A (en) | 2020-11-02 | 2020-11-02 | YOLOv 4-based multi-target vehicle detection and tracking method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113205108A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113702393A (en) * | 2021-09-29 | 2021-11-26 | 安徽理工大学 | Intrinsic safety type mining conveyor belt surface damage detection system and detection method |
CN116993779A (en) * | 2023-08-03 | 2023-11-03 | 重庆大学 | Vehicle target tracking method suitable for monitoring video |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110378331A (en) * | 2019-06-10 | 2019-10-25 | 南京邮电大学 | A kind of end-to-end Vehicle License Plate Recognition System and its method based on deep learning |
CN111126152A (en) * | 2019-11-25 | 2020-05-08 | 国网信通亿力科技有限责任公司 | Video-based multi-target pedestrian detection and tracking method |
CN111259819A (en) * | 2020-01-16 | 2020-06-09 | 广东工业大学 | Outdoor scene safety monitoring method based on visual correlation discrimination network |
CN111476826A (en) * | 2020-04-10 | 2020-07-31 | 电子科技大学 | Multi-target vehicle tracking method based on SSD target detection |
CN111488795A (en) * | 2020-03-09 | 2020-08-04 | 天津大学 | Real-time pedestrian tracking method applied to unmanned vehicle |
CN111832513A (en) * | 2020-07-21 | 2020-10-27 | 西安电子科技大学 | Real-time football target detection method based on neural network |
-
2020
- 2020-11-02 CN CN202011206816.1A patent/CN113205108A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110378331A (en) * | 2019-06-10 | 2019-10-25 | 南京邮电大学 | A kind of end-to-end Vehicle License Plate Recognition System and its method based on deep learning |
CN111126152A (en) * | 2019-11-25 | 2020-05-08 | 国网信通亿力科技有限责任公司 | Video-based multi-target pedestrian detection and tracking method |
CN111259819A (en) * | 2020-01-16 | 2020-06-09 | 广东工业大学 | Outdoor scene safety monitoring method based on visual correlation discrimination network |
CN111488795A (en) * | 2020-03-09 | 2020-08-04 | 天津大学 | Real-time pedestrian tracking method applied to unmanned vehicle |
CN111476826A (en) * | 2020-04-10 | 2020-07-31 | 电子科技大学 | Multi-target vehicle tracking method based on SSD target detection |
CN111832513A (en) * | 2020-07-21 | 2020-10-27 | 西安电子科技大学 | Real-time football target detection method based on neural network |
Non-Patent Citations (4)
Title |
---|
ALEXEY BOCHKOVSKIY: "YOLOv4: Optimal Speed and Accuracy of Object Detection", 《COMPUTER SCIENCE》 * |
CSDN博客园作者"S.ZHENKAI": "yolov3改进4层特征检测层", 《HTTPS://BLOG.CSDN.NET/WEIXIN_44076342/ARTICLE/DETAILS/106547312》 * |
徐子睿: "基于YOLOv4的车辆检测与流量统计研究", 《现代信息科技》 * |
码农家园: "YOLOV4中k-means聚类获得anchor boxes", 《HTTPS://WWW.CODENONG.COM/CS109071574/》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113702393A (en) * | 2021-09-29 | 2021-11-26 | 安徽理工大学 | Intrinsic safety type mining conveyor belt surface damage detection system and detection method |
CN113702393B (en) * | 2021-09-29 | 2023-10-27 | 安徽理工大学 | Intrinsic safety type mining conveyor belt surface damage detection system and detection method |
CN116993779A (en) * | 2023-08-03 | 2023-11-03 | 重庆大学 | Vehicle target tracking method suitable for monitoring video |
CN116993779B (en) * | 2023-08-03 | 2024-05-14 | 重庆大学 | Vehicle target tracking method suitable for monitoring video |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111680542B (en) | Steel coil point cloud identification and classification method based on multi-scale feature extraction and Pointnet neural network | |
CN112308881B (en) | Ship multi-target tracking method based on remote sensing image | |
CN108573496B (en) | Multi-target tracking method based on LSTM network and deep reinforcement learning | |
CN107424171B (en) | Block-based anti-occlusion target tracking method | |
CN111476826A (en) | Multi-target vehicle tracking method based on SSD target detection | |
CN110197502B (en) | Multi-target tracking method and system based on identity re-identification | |
CN107633226B (en) | Human body motion tracking feature processing method | |
CN111444767B (en) | Pedestrian detection and tracking method based on laser radar | |
CN112052802B (en) | Machine vision-based front vehicle behavior recognition method | |
CN111862155B (en) | Unmanned aerial vehicle single vision target tracking method aiming at target shielding | |
CN110532921B (en) | SSD-based generalized label detection multi-Bernoulli video multi-target tracking method | |
CN112288773A (en) | Multi-scale human body tracking method and device based on Soft-NMS | |
CN111739053B (en) | Online multi-pedestrian detection tracking method under complex scene | |
CN113327272B (en) | Robustness long-time tracking method based on correlation filtering | |
CN110363165B (en) | Multi-target tracking method and device based on TSK fuzzy system and storage medium | |
CN111582349B (en) | Improved target tracking algorithm based on YOLOv3 and kernel correlation filtering | |
CN112016445A (en) | Monitoring video-based remnant detection method | |
CN113205108A (en) | YOLOv 4-based multi-target vehicle detection and tracking method | |
CN104615998B (en) | A kind of vehicle retrieval method based on various visual angles | |
CN109146918B (en) | Self-adaptive related target positioning method based on block | |
CN112946625B (en) | B-spline shape-based multi-extended target track tracking and classifying method | |
CN110400347B (en) | Target tracking method for judging occlusion and target relocation | |
Cao et al. | Correlation-based tracking of multiple targets with hierarchical layered structure | |
KR101460313B1 (en) | Apparatus and method for robot localization using visual feature and geometric constraints | |
CN116381672A (en) | X-band multi-expansion target self-adaptive tracking method based on twin network radar |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210803 |
|
WD01 | Invention patent application deemed withdrawn after publication |