CN113763427B - Multi-target tracking method based on coarse-to-fine shielding processing - Google Patents
Multi-target tracking method based on coarse-to-fine shielding processing Download PDFInfo
- Publication number
- CN113763427B CN113763427B CN202111035065.6A CN202111035065A CN113763427B CN 113763427 B CN113763427 B CN 113763427B CN 202111035065 A CN202111035065 A CN 202111035065A CN 113763427 B CN113763427 B CN 113763427B
- Authority
- CN
- China
- Prior art keywords
- shielding
- pedestrian
- target
- model
- pedestrians
- 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
- 238000012545 processing Methods 0.000 title claims abstract description 31
- 238000000034 method Methods 0.000 title claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 27
- 239000013598 vector Substances 0.000 claims abstract description 24
- 238000001514 detection method Methods 0.000 claims abstract description 19
- 238000000605 extraction Methods 0.000 claims abstract description 5
- 238000012544 monitoring process Methods 0.000 claims description 8
- 238000013507 mapping Methods 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 7
- 238000005259 measurement Methods 0.000 abstract description 3
- 208000006440 Open Bite Diseases 0.000 abstract 1
- 230000006870 function Effects 0.000 description 12
- 238000010586 diagram Methods 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000001629 suppression Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 230000008014 freezing Effects 0.000 description 2
- 238000007710 freezing Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses a multi-target tracking method based on coarse-to-fine shielding treatment, which comprises the following steps of: firstly, adding an occlusion score prediction on the basis of a JDE model pre-measurement head, thereby completing the complete processing of a non-occlusion target and the rough processing of the occlusion target; on the basis, training the mapped and cut area for shielding pedestrians as a training set of the second step model, and finishing accurate detection and apparent feature vector extraction for shielding pedestrians; thereby realizing the fine treatment of the shielding target; and integrating the results output by the two-step model, and completing the tracking of pedestrians by using a data association algorithm. The invention solves the problem that the prior art can not accurately track pedestrians in a scene with shielding condition, and can be well adapted to public environments with various time periods and various pedestrian densities; the pedestrian tracking method has a good effect on pedestrian tracking.
Description
Technical Field
The invention belongs to the field of computer vision and monitoring video analysis, and particularly relates to a multi-target tracking method based on coarse-to-fine shielding processing.
Background
Multi-objective tracking is an important component of surveillance video analysis. The method can be directly used for analyzing the motion trail of the object, and can be used as a research basis for advanced tasks such as object motion recognition, behavior analysis and the like.
In order to accomplish the multi-objective tracking task, many mainstream deep learning algorithms propose strategies for tracking based on detection. These methods divide multi-target tracking into a detection module and an embedding module. The detection module completes target detection, the embedding module extracts the characteristics of the target by using a related algorithm, however, repeated calculation can occur between the two modules for a plurality of times, and the running speed is influenced. For this reason, some scientists have proposed a method of integrating a detection module and an embedding module into one neural network, the two modules sharing the same underlying characteristics, thereby avoiding repeated calculations and improving performance. However, due to limitations of the self-detection framework, these methods do not perform well in some scenarios for detection and tracking of occluding objects. In particular, the detection framework often detects two occluded objects as one object, which also presents some problems for object tracking. For occlusion target detection problems, some improved non-maximum suppression algorithms are proposed, such as soft-NMS, adaptive-NMS, etc., and some improved loss functions are proposed, such as retransmission loss, etc., but these methods all depend on the predicted results of the original detection network, and for some scenarios, the occlusion detection is not good.
Disclosure of Invention
In order to solve the problems, the invention discloses a multi-target tracking method based on coarse-to-fine shielding treatment, which separates the treatment of a non-shielding target from a shielding target, and in a first step model, focuses on the non-shielding target and positions the shielding target to realize the coarse treatment of the shielding target; and the second step of model realizes the fine treatment of the shielding target, thereby realizing the improvement of the pedestrian tracking performance in some shielding environments.
In order to achieve the above purpose, the technical scheme of the invention is as follows:
a multi-target tracking method based on coarse-to-fine occlusion processing, comprising the steps of:
firstly, labeling the data set obtained from public places, constructing a training set and a testing set for pedestrian tracking,
training a first step model of the multi-target tracking method by utilizing part of information marked by the training set to realize detection of non-shielding pedestrian targets, extraction of corresponding apparent feature vectors and positioning of shielding pedestrian areas, thereby completing complete processing of the non-shielding targets and rough processing of the shielding targets,
and thirdly, mapping the area for shielding the pedestrians on the basis of the training completion of the first step model, cutting out the area as a training set of the second step model for training, and completing detection and apparent feature vector extraction of the shielded pedestrians. Thereby realizing the fine treatment of the shielding target,
and step four, integrating the results output by the two-step model, and completing the tracking of pedestrians by using a data association algorithm.
In the first step, the pedestrian tracking dataset is constructed on the monitoring videos shot by the cameras in the public places, the labeling content comprises a pedestrian bounding box id (marked as 0 when no shielding condition exists and increased id when 2 persons are shielded mutually), the pedestrian id (marked as-1 when shielding condition exists and increased from 0 otherwise), the bounding box position, whether shielding exists for the pedestrian in the bounding box (shielding mark 1 and non-shielding mark 0), and the pedestrian target is contained in which bounding box (marked as 0 when shielding condition exists and marked as shielding bounding box id containing the pedestrian target when no shielding condition exists);
further, in the second step, the first step model of the method defines two kinds of bounding box regression, one is that of a single pedestrian without shielding, and the other is that of two pedestrians with shielding; on the basis of the JDE model, the first step model adds shielding fraction prediction in a prediction head so that the prediction head can judge whether a pedestrian in a regressed bounding box is shielded or not; if no shielding exists, the position and the apparent feature vector of the pedestrian can be extracted at the same time, and if the shielding exists, the shielding area is positioned, so that the rough treatment of shielding the pedestrian is completed.
In the third step, the occlusion region located in the second step is mapped to the small-scale feature map of the first step model, and is cut by using an ROI alignment algorithm as an input of the second step model, the second step model performs fine processing on the occlusion region feature map to obtain a position and an apparent feature vector of an occluded pedestrian, wherein when the boundary box of the occluded pedestrian is trained, a loss function adopts a mode of combining SmoothL1 loss and Repgt and Repbox loss weighting, and the boundary box loss function is calculated as follows:
wherein,representing smoothL1 loss function, L RepGT Representing the RepGT loss function, L RepBox Indicating the RepBox loss function, and α, β indicating the weighting parameters.
Further, in the fourth step, the results in the first step and the second step are combined to obtain the positions and apparent feature vectors of all the pedestrian targets in the current frame, and the characteristics of similarity and small position variation of the same pedestrian target apparent feature vector between adjacent frames are utilized to complete the matching of the pedestrian targets, so that the tracking of the multi-target pedestrians is finally completed.
The beneficial effects of the invention are as follows:
1) The invention creatively provides a coarse-to-fine shielding processing method, which comprises the steps of firstly completing coarse processing of shielding targets and complete processing of non-shielding targets, secondly completing fine processing of shielding targets, and finally comprehensively obtaining final pedestrian positions and corresponding apparent feature vectors by the results of the two steps.
2) According to the first-step model, based on the JDE model, the shielding score prediction is added in the prediction head, so that whether the shielding condition exists in the bounding box output by the first-step model can be judged, and if the shielding condition exists, the shielding bounding box can be positioned, and the rough processing of the shielding target is realized.
3) The invention maps the shielding pedestrian boundary box positioned by the first step model onto the small-scale feature map of the first step model (the small-scale feature map retains more information) and adopts the ROI alignment algorithm to cut, and the shielding pedestrian boundary box is used as the input of the shielding fine processing model (the second step model).
4) The invention designs a network structure of the shielding fine processing model (a second step model), and adopts a mode of combining SmoothL1 loss with RepGT and RepBox loss weighting as a loss function of shielding fine processing model boundary frame regression.
5) The invention provides a two-stage model training method aiming at a two-stage model framework, wherein in the first stage, parameters of a second-stage model are frozen, and only a first-stage model is trained; and in the second stage, on the basis of the completion of the training of the first step model, freezing parameters of the first model, and training by taking a feature map corresponding to the shielding pedestrian boundary frame positioned by the first step model as the input of the second step model. The problem that pedestrians in a scene with shielding conditions cannot be accurately tracked in the prior art is solved, and the system is well suitable for public environments with various periods and various pedestrian densities; the pedestrian tracking method has a good effect on pedestrian tracking.
Drawings
Fig. 1 is a flow chart of the present invention.
Fig. 2 is a scene diagram under a subway monitor video.
FIG. 3 is a diagram of a multi-target tracking network model framework based on coarse-to-fine occlusion processing.
FIG. 4 is a schematic diagram of a first step model predictive head and loss function.
Fig. 5 is a network configuration diagram of the occlusion finishing model (second-step model).
Fig. 6 is a multi-target tracking effect display diagram.
Detailed Description
The present invention is further illustrated in the following drawings and detailed description, which are to be understood as being merely illustrative of the invention and not limiting the scope of the invention.
Examples
The monitoring video used in the model training of this embodiment is a monitoring video in an actual subway scene, and a scene diagram is shown in fig. 2.
In this embodiment, taking the subway station monitoring video shown in fig. 2 as an example, these video images include both pedestrians in the case where there is no shielding and pedestrians in the case where there is shielding. After the video in the subway scene is obtained, marking pedestrians in the video, and thus obtaining a subway pedestrian multi-target tracking data set.
The embodiment provides a multi-target tracking method based on coarse-to-fine shielding processing, which is to perform coarse-to-fine model processing on shielding targets in an input image, and achieve better multi-target tracking effect by combining modes such as optimizing a loss function, optimizing NMS (network management system) and the like on the basis, wherein a frame diagram of the model is shown in figure 3, and the specific steps are as follows:
1) Image input processing: to eliminate the adverse effect of the input image size on model training and to take into account factors such as resolution, each frame of image in the video is resized to 1088 x 608 before being input to the model.
2) Adding occlusion score prediction: the first step model adds shielding fraction prediction into the pre-measurement head based on the JDE model, and the pre-measurement head comprises four prediction values of confidence prediction, bounding box regression prediction, shielding fraction prediction and apparent feature vector; the confidence level, the shielding score and the apparent feature vector are trained by cross entropy loss, and the boundary box regression is trained by smoothL1 loss. The weighting parameters of the loss values of all parts in the loss function are determined in an adaptive training mode. The loss function calculation formula is as follows:
wherein M is the number of the pre-measuring heads,representing confidence loss, occlusion prediction loss, bounding box regression loss and apparent feature vector loss, respectively,/>For the weighted parameters associated with each penalty, a learnable parameter is modeled. The schematic diagram is shown in fig. 4:
3) Feature map mapping and clipping: after the first step model locates the shielding boundary frame, mapping the shielding boundary frame after non-maximum suppression to a small-scale feature map of the first step model, cutting by adopting a ROIAlign algorithm, and taking the cut feature map as the input of the second step model.
4) Fine treatment of shielding targets: the purpose of the occlusion finishing model (second step model) is to accurately acquire the bounding box and corresponding apparent feature vector of the occluding pedestrian. The network structure is shown in fig. 5, the cut feature map has two branches after passing through a convolution block, one branch is used for confidence prediction and regression of a pedestrian boundary box, and the other branch is used for extracting apparent feature vectors corresponding to the blocked pedestrians.
5) Loss function of occlusion finishing model: confidence and apparent feature vector training of the occlusion finishing model still adopts cross entropy loss, and training of bounding box regression adopts a mode of combining SmoothL1 loss with RepGT and RepBox loss weighting. The purpose of SmoothL1 loss is to make the bounding box predicted value as close to the true value as possible, and the design of the repagt loss is to keep the predicted bounding box as far away as possible from its neighboring true bounding box. The purpose of the RepBox penalty is to keep the two bounding box predictors returning to different bounding boxes as far apart as possible. These three bounding box losses coordinate with each other, having a positive effect on regression of the bounding box that obscures the pedestrian. Meanwhile, the bounding box regresses the loss, the confidence coefficient is lost, and the weighting mode of the apparent feature vector loss is the same as that of the loss value of the first step model.
6) Integration and optimization: for the bounding box output by the second step model (occlusion finishing model), firstly, removing some bounding boxes which do not meet the requirements, such as the fact that the area is too small, and the coordinates of the bounding boxes exceed the image area; meanwhile, after non-maximum suppression processing is carried out on the bounding box output by the second step model by using the Soft-NMS algorithm, the bounding box output by the first step model is integrated, and then the final result is obtained by carrying out Soft-NMS algorithm processing again.
7) Data association and trajectory generation: the method comprises the steps of firstly taking cosine distance between an apparent characteristic vector group of a detection target group and an apparent characteristic vector group of a tracking target group as a main matching principle, taking distance between a boundary frame of the detection target group and a boundary frame predicted by a Kalman filter for the tracking target group as an auxiliary matching principle, respectively selecting weighting ratios of 0.95 and 0.05, and adopting a Hungary matching algorithm to finish first-step matching; secondly, in the second step of matching, the intersection ratio (IOU) of the boundary box of the detection target group and the boundary box of the tracking target group is used as a matching principle, and still the Hungary algorithm is adopted to complete the matching. For the previously unmatched tracking target marked as lost, the lost tracking target remains engaged in the matching of the next frame, and when the lost tracking target has not been matched to a new target for 25 consecutive frames (herein, frame rate is a threshold herein), this tracking target is removed from the population of tracking targets and no longer engaged in the matching of the tracking target with the detected target.
8) Two-stage model training: freezing the parameters of the second step model when the first step model is trained, and only training the first step model; after the training of the first step model is completed, the parameters of the first step model are frozen, and the feature map corresponding to the boundary box of the blocked pedestrian positioned by the first step model is used as the input of the second step model for training.
After the model is built and trained, the pedestrian in the actual monitoring video can be tracked, the tracking effect is shown in fig. 6, and the interval between two adjacent pictures is 25 frames.
The invention provides a multi-target tracking method based on coarse-to-fine shielding processing. The model is divided into two steps, wherein the first step model is used for realizing the full processing (positioning) of the non-shielding target and the rough processing (positioning) of the shielding target; and the second step of model is used for accurately positioning the blocked pedestrians and extracting apparent feature vectors thereof, so that the precise treatment of the blocking target is realized. And finally, integrating and optimizing the output results of the two models, and then carrying out data association and track generation to finally finish the multi-target tracking task. The invention has important effect on the monitoring video analysis task in the field of computer vision and has a large application prospect.
The present invention is not limited to the specific technical solutions described in the above embodiments, and other embodiments may be used in addition to the above embodiments. Any modifications, equivalent substitutions, improvements, etc. made by those skilled in the art, which are within the spirit and principles of the present invention, are intended to be included within the scope of the present invention.
Claims (3)
1. A multi-target tracking method based on coarse-to-fine occlusion processing, the method comprising the steps of:
firstly, marking a data set obtained from a public place, and constructing a training set and a testing set for pedestrian tracking;
training a first step model of the multi-target tracking method by utilizing part of information marked by the training set; the detection of non-shielding pedestrian targets and the extraction of corresponding apparent feature vectors are realized, and the positioning of shielding pedestrian areas is realized; thus completing the complete processing of the non-shielding target and the rough processing of the shielding target;
the first step model of the method defines two kinds of bounding box regression, one is that of single pedestrian without shielding, and the other is that of two pedestrians with shielding; on the basis of the JDE model, the first step model adds shielding fraction prediction in a prediction head so that the prediction head can judge whether a pedestrian in a regressed bounding box is shielded or not; if no shielding exists, the position and the apparent feature vector of the pedestrian can be extracted at the same time, and if the shielding exists, a shielding area is positioned, so that the rough treatment of shielding the pedestrian is completed;
training the mapped and cut area for shielding pedestrians as a training set of a second step model on the basis of the model training, and finishing accurate detection and apparent feature vector extraction for shielding pedestrians; thereby realizing the fine treatment of the shielding target;
mapping the occlusion region positioned in the second step into a small-scale feature map of the first step model, cutting by adopting an ROI alignment algorithm to serve as input of the second step model, and performing fine processing on the occlusion region feature map by the second step model to obtain the position and apparent feature vector of the occluded pedestrian, wherein when the boundary box of the occluded pedestrian is trained, a loss function adopts a mode of combining smoothL1 loss with RepGT and RepBox loss weighting, and the boundary box loss function is calculated as follows:
wherein,representing smoothL1 loss function, L RepGT Representing the RepGT loss function, L RepBox Representing the RepBox loss function, alpha, beta representing the weighting parameters;
and step four, integrating the results output by the two-step model, and completing the tracking of pedestrians by using a data association algorithm.
2. A multi-target tracking method based on coarse-to-fine occlusion processing as claimed in claim 1, wherein: in the first step, a pedestrian tracking data set is constructed on monitoring videos shot by cameras in a plurality of public places, and labeling contents comprise a pedestrian bounding box id, a pedestrian id, a bounding box position, whether a pedestrian is shielded in the bounding box or not, and in which bounding box a pedestrian target is contained, wherein the information forms a truth value label in a training set and a testing set.
3. A multi-target tracking method based on coarse-to-fine occlusion processing as claimed in claim 1, wherein: and in the fourth step, the results in the first step and the second step are combined to obtain the positions and apparent feature vectors of all pedestrian targets in the current frame, and the characteristics of similarity and small position variation of the same apparent feature vector of the pedestrian target between adjacent frames are utilized to complete the matching of the pedestrian targets, so that the task is finally completed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111035065.6A CN113763427B (en) | 2021-09-05 | 2021-09-05 | Multi-target tracking method based on coarse-to-fine shielding processing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111035065.6A CN113763427B (en) | 2021-09-05 | 2021-09-05 | Multi-target tracking method based on coarse-to-fine shielding processing |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113763427A CN113763427A (en) | 2021-12-07 |
CN113763427B true CN113763427B (en) | 2024-02-23 |
Family
ID=78792988
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111035065.6A Active CN113763427B (en) | 2021-09-05 | 2021-09-05 | Multi-target tracking method based on coarse-to-fine shielding processing |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113763427B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114120373A (en) * | 2022-01-24 | 2022-03-01 | 苏州浪潮智能科技有限公司 | Model training method, device, equipment and storage medium |
WO2023197232A1 (en) * | 2022-04-14 | 2023-10-19 | 京东方科技集团股份有限公司 | Target tracking method and apparatus, electronic device, and computer readable medium |
CN116129432B (en) * | 2023-04-12 | 2023-06-16 | 成都睿瞳科技有限责任公司 | Multi-target tracking labeling method, system and storage medium based on image recognition |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112836639A (en) * | 2021-02-03 | 2021-05-25 | 江南大学 | Pedestrian multi-target tracking video identification method based on improved YOLOv3 model |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11055854B2 (en) * | 2018-08-23 | 2021-07-06 | Seoul National University R&Db Foundation | Method and system for real-time target tracking based on deep learning |
-
2021
- 2021-09-05 CN CN202111035065.6A patent/CN113763427B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112836639A (en) * | 2021-02-03 | 2021-05-25 | 江南大学 | Pedestrian multi-target tracking video identification method based on improved YOLOv3 model |
Non-Patent Citations (1)
Title |
---|
基于YOLOv3与卡尔曼滤波的多目标跟踪算法;任珈民;宫宁生;韩镇阳;;计算机应用与软件(第05期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN113763427A (en) | 2021-12-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113763427B (en) | Multi-target tracking method based on coarse-to-fine shielding processing | |
EP1836683B1 (en) | Method for tracking moving object in video acquired of scene with camera | |
Zhou et al. | Efficient road detection and tracking for unmanned aerial vehicle | |
US20230289979A1 (en) | A method for video moving object detection based on relative statistical characteristics of image pixels | |
She et al. | Vehicle tracking using on-line fusion of color and shape features | |
CN101957997A (en) | Regional average value kernel density estimation-based moving target detecting method in dynamic scene | |
Shukla et al. | Moving object tracking of vehicle detection: a concise review | |
CN101916446A (en) | Gray level target tracking algorithm based on marginal information and mean shift | |
CN105321189A (en) | Complex environment target tracking method based on continuous adaptive mean shift multi-feature fusion | |
Tang et al. | Multiple-kernel adaptive segmentation and tracking (MAST) for robust object tracking | |
Hu et al. | A novel approach for crowd video monitoring of subway platforms | |
CN113706584A (en) | Streetscape flow information acquisition method based on computer vision | |
Roy et al. | A comprehensive survey on computer vision based approaches for moving object detection | |
CN116012949B (en) | People flow statistics and identification method and system under complex scene | |
CN111986233B (en) | Large-scene minimum target remote sensing video tracking method based on feature self-learning | |
Shbib et al. | Distributed monitoring system based on weighted data fusing model | |
CN110322474B (en) | Image moving target real-time detection method based on unmanned aerial vehicle platform | |
Davies et al. | Using CART to segment road images | |
Streib et al. | Extracting Pathlets FromWeak Tracking Data | |
Chanawangsa et al. | A new color-based lane detection via Gaussian radial basis function networks | |
Bhuvaneshwar et al. | Real-time detection of crossing pedestrians for traffic-adaptive signal control | |
Pulare et al. | Implementation of real time multiple object detection and classification of HEVC videos | |
Chen et al. | Pedestrian tracking algorithm based on Kalman filter and partial mean-shift tracking | |
Zhang et al. | Research on single object tracking algorithm based on Siamese network and Kalman filter | |
Sanap et al. | Survey on moving object detection |
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 |