CN117710903B - Visual specific pedestrian tracking method and system based on ReID and Yolov5 double models - Google Patents
Visual specific pedestrian tracking method and system based on ReID and Yolov5 double models Download PDFInfo
- Publication number
- CN117710903B CN117710903B CN202410161198.5A CN202410161198A CN117710903B CN 117710903 B CN117710903 B CN 117710903B CN 202410161198 A CN202410161198 A CN 202410161198A CN 117710903 B CN117710903 B CN 117710903B
- Authority
- CN
- China
- Prior art keywords
- picture
- loss
- model
- distance
- extracting
- 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 48
- 230000000007 visual effect Effects 0.000 title claims abstract description 14
- 238000012549 training Methods 0.000 claims abstract description 34
- 238000000605 extraction Methods 0.000 claims abstract description 10
- 238000001514 detection method Methods 0.000 claims abstract description 8
- 238000012360 testing method Methods 0.000 claims abstract description 7
- 238000013526 transfer learning Methods 0.000 claims abstract description 7
- 238000002372 labelling Methods 0.000 claims description 18
- 238000007781 pre-processing Methods 0.000 claims description 9
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 6
- 230000006870 function Effects 0.000 abstract description 2
- 238000010606 normalization Methods 0.000 abstract description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a visual specific pedestrian tracking method and system based on ReID and Yolov double models, belonging to the technical field of image recognition, wherein the method comprises the following steps: (1) Pedestrian data are obtained, and the data are preprocessed and stored; (2) Extracting the preprocessed data frame by frame, extracting person pictures in each frame of pictures by YOLOv, and storing the person pictures; (3) Acquiring a mark 1501 data set, randomly dividing a test set and a training set, and training Resnet a model to acquire a final model; (4) Inputting the trained model into the picture obtained in the step (2) through a transfer learning method, and extracting the characteristics; (5) Combining the extracted pictures into a tensor, transmitting the tensor to a ReID model, and carrying out feature extraction and normalization; (6) determining the magnitude of the minimum average distance and the distance threshold; (7) saving the final video; according to the invention, by means of YOLOv target detection of a specific class and ReID search of a specific ID, the function of tracking offender in collaborative search is realized.
Description
Technical Field
The invention relates to the technical field of image recognition, in particular to a visual specific pedestrian tracking method and system based on ReID and Yolov5 double models.
Background
In recent years, with the great increase of the storage amount of non-motor vehicles in campuses and the improvement of the opening degree of universities, more and more non-motor vehicles enter the campuses and illegal parking occurs. The safety problem of campus traffic is increasingly outstanding, and the safety problem of campus traffic is deeply concerned by school leaders, teachers and students. Therefore, how to solve the safety problem of campus traffic is urgent.
Disclosure of Invention
The invention aims to: the invention aims to provide a visual specific pedestrian tracking method and system based on ReID and Yolov double models, which realize pedestrian re-identification of photo-video under the double models by means of model transformation through a self-training model of a mark 1501 dataset and a strong base line ReID and a Yolov5s.pt weight file initialization model, and search specific IDs frame by frame to solve the security problem of a campus.
The technical scheme is as follows: the invention discloses a visual specific pedestrian tracking method based on ReID and Yolov double models, which comprises the following steps:
(1) Acquiring pedestrian data by using a plurality of 1280 x 1080 high-definition cameras and SD cameras, preprocessing the data and storing the data;
(2) Extracting the preprocessed data frame by frame, extracting a person picture in each frame of picture by YOLOv, and storing the picture by extracting the person picture coordinates;
(3) Acquiring a mark 1501 data set, randomly dividing a test set and a training set, and training Resnet a model to acquire a final model;
(4) Inputting the trained model into the picture obtained in the step (2) through a transfer learning method, and extracting the characteristics;
(5) Combining the extracted pictures into a tensor, transmitting the tensor to a trained ReID model, extracting and normalizing the characteristics, and finding out the index of the minimum average distance by calculating the mean value of Euclidean distances between the query picture and each stored picture;
(6) Setting corresponding distance thresholds through different scenes, judging the size of the minimum average distance and the distance threshold, and outputting a final target boundary frame when the minimum average distance is smaller than the distance threshold, wherein the final target boundary frame comprises position coordinates, confidence coefficient and category information of the boundary frame;
(7) And saving the final video.
Further, in the step (1), the pretreatment is specifically as follows: labeling the data, including manual labeling and labeling with a DPM detector; fixing the image size to 256×128; each image is segmented with a probability level of 0.5, each image is decoded into 32-bit anchor point original pixel values, and the RGB channels are normalized.
Further, in the step (1), the preprocessed data is stored, and naming criteria are as follows: 000N_cNsN_000ABC_0N.jpeg or 000N_cNsN_000ABC_00.jpeg; wherein 000N represents the tag number of each person; cN represents the nth camera; sN represents an nth video clip; 000ABC represents the picture of the 000 th ABC frame of cNsN; 0N represents the Nth detection box on cNsN _000ABC frames.
Further, the step (3) specifically includes the following steps: the dimension of the Resnet model full-connection layer is modified to N; optimizing the model by adopting an Adam method; selecting identity loss, center loss and enhanced triplet loss as total losses in back propagation; the formula is as follows:
；
Wherein, For identity loss,/>For enhanced triplet loss,/>Is the center loss.
Further, the identity loss formula is as follows:
；
wherein n represents the number of training samples in each batch, provided with a given label Input image/>ThenRepresentation/>Identified as category/>Is used for predicting the probability of (1);
The enhanced triplet loss is as follows:
；
wherein (i, j, k) represents each anchor sample Triplet in each training batch,/>Representing the distance between positive sample pairs,/>Representing the distance between the negative pair of samples; for anchor samples/>And/>Is the corresponding positive sample; /(I)Expressed as anchor sample/>And its corresponding positive sample/>Weight value of Euclidean distance of/>Is a weight value, N represents the positive sample pair number, p represents the environment of the positive sample pair, and N represents the environment of the negative sample pair;
the euclidean distance between two samples is:
；
Wherein, And/>Representation/>And/>Corresponding feature vectors;
the definition formula of k is:
；
the square error is expressed as:
；
The center loss formula is as follows:
；
Wherein, Representing identity/>Is a class center of (c).
The invention discloses a visual specific pedestrian tracking system based on ReID and Yolov double models, which comprises the following components:
and a pretreatment module: the method comprises the steps that pedestrian data are obtained by using a plurality of 1280 x 1080 high-definition cameras and SD cameras, and are preprocessed and stored;
And a picture extraction module: the method comprises the steps of extracting preprocessed data frame by frame, extracting a person picture in each frame of picture by YOLOv, and storing the picture by extracting a person picture coordinate;
Model training module: the method comprises the steps of obtaining a mark 1501 data set, randomly dividing a test set and a training set, training Resnet models, and obtaining a final model;
And the feature extraction module is used for: the method comprises the steps of inputting a trained model into a picture obtained by a picture extraction module through a transfer learning method, and extracting features;
And an index module: the method comprises the steps of merging extracted pictures into a tensor, transmitting the tensor to a trained ReID model, extracting and normalizing characteristics, and finding out an index of the minimum average distance by calculating the average value of Euclidean distances between a query picture and each stored picture;
And an output module: the method comprises the steps of setting corresponding distance thresholds through different scenes, judging the size of the minimum average distance and the distance threshold, and outputting a final target boundary frame when the minimum average distance is smaller than the distance threshold, wherein the final target boundary frame comprises position coordinates, confidence and category information of the boundary frame;
And a storage module: for saving the final video.
Further, in the preprocessing module, the preprocessing is specifically as follows: labeling the data, including manual labeling and labeling with a DPM detector; fixing the image size to 256×128; each image is segmented with a probability level of 0.5, each image is decoded into 32-bit anchor point original pixel values, and the RGB channels are normalized.
Further, in the preprocessing module, the preprocessed data is stored, and naming criteria are as follows: 000N_cNsN_000ABC_0N.jpeg or 000N_cNsN_000ABC_00.jpeg; wherein 000N represents the tag number of each person; cN represents the nth camera; sN represents an nth video clip; 000ABC represents the picture of the 000 th ABC frame of cNsN; 0N represents the Nth detection box on cNsN _000ABC frames.
Further, in the model training module, the specific steps are as follows: the dimension of the Resnet model full-connection layer is modified to N; optimizing the model by adopting an Adam method; selecting identity loss, center loss and enhanced triplet loss as total losses in back propagation; the formula is as follows:
；
Wherein, For identity loss,/>For enhanced triplet loss,/>Is the center loss.
Further, in the model training module, the identity loss formula is as follows:
；
wherein n represents the number of training samples in each batch, provided with a given label Input image/>ThenRepresentation/>Identified as category/>Is used for predicting the probability of (1);
The enhanced triplet loss is as follows:
；
wherein (i, j, k) represents each anchor sample Triplet in each training batch,/>Representing the distance between positive sample pairs,/>Representing the distance between the negative pair of samples; for anchor samples/>And/>Is the corresponding positive sample; /(I)Expressed as anchor sample/>And its corresponding positive sample/>Weight value of Euclidean distance of/>Is a weight value, N represents the positive sample pair number, p represents the environment of the positive sample pair, and N represents the environment of the negative sample pair;
the euclidean distance between two samples is:
；
Wherein, And/>Representation/>And/>Corresponding feature vectors;
the definition formula of k is:
；
the square error is expressed as:
；
The center loss formula is as follows:
；
Wherein, Representing identity/>Is a class center of (c).
The beneficial effects are that: compared with the prior art, the invention has the following remarkable advantages: the function of cooperatively searching and tracking the offender under multiple cameras is realized by YOLOv for target detection of a specific class (person) and ReID for searching of a specific ID.
Drawings
Fig. 1 is a schematic diagram of the principle of the present invention.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings.
As shown in fig. 1, an embodiment of the present invention provides a method for tracking a specific pedestrian based on ReID and Yolov5 double models, which includes the following steps:
(1) Acquiring pedestrian data by using a plurality of 1280 x 1080 high-definition cameras and SD cameras, preprocessing the data and storing the data; the pretreatment is specifically as follows: labeling the data, including manual labeling and labeling with a DPM detector; fixing the image size to 256×128; dividing each image with 0.5 probability level, decoding each image into 32-bit locating point original pixel values, and carrying out normalization processing on RGB channels; storing the preprocessed data, wherein naming criteria are as follows: 000N_cNsN_000ABC_0N.jpeg or 000N_cNsN_000ABC_00.jpeg; wherein 000N represents the tag number of each person; cN represents the nth camera; sN represents an nth video clip; 000ABC represents the picture of the 000 th ABC frame of cNsN; 0N represents the Nth detection box on cNsN _000ABC frame
(2) Extracting the preprocessed data frame by frame, extracting a person picture in each frame of picture by YOLOv, and storing the picture by extracting the person picture coordinates;
(3) Acquiring a mark 1501 data set, randomly dividing a test set and a training set, and training Resnet a model to acquire a final model; the method comprises the following steps: the dimension of the Resnet model full-connection layer is modified to N; optimizing the model by adopting an Adam method; selecting identity loss, center loss and enhanced triplet loss as total losses in back propagation; the formula is as follows:
；
Wherein, For identity loss,/>For enhanced triplet loss,/>Is the center loss.
The identity loss formula is as follows:
；
wherein n represents the number of training samples in each batch, provided with a given label Input image/>ThenRepresentation/>Identified as category/>Is used for predicting the probability of (1);
The enhanced triplet loss is as follows:
；
wherein (i, j, k) represents each anchor sample Triplet in each training batch,/>Representing the distance between positive sample pairs,/>Representing the distance between the negative pair of samples; for anchor samples/>And/>Is the corresponding positive sample; /(I)Expressed as anchor sample/>And its corresponding positive sample/>Weight value of Euclidean distance of/>Is a weight value, N represents the positive sample pair number, p represents the environment of the positive sample pair, and N represents the environment of the negative sample pair;
the euclidean distance between two samples is:
；
Wherein, And/>Representation/>And/>Corresponding feature vectors;
the definition formula of k is:
；
the square error is expressed as:
；
The center loss formula is as follows:
；
Wherein, Representing identity/>Is a class center of (c).
(4) Inputting the trained model into the picture obtained in the step (2) through a transfer learning method, and extracting the characteristics;
(5) And combining the extracted pictures into a tensor, transmitting the tensor to a trained ReID model, extracting and normalizing the characteristics, and finding the index of the minimum average distance by calculating the mean value of Euclidean distances between the query picture and each stored picture.
(6) Setting corresponding distance thresholds through different scenes, judging the size of the minimum average distance and the distance threshold, and outputting a final target boundary frame when the minimum average distance is smaller than the distance threshold, wherein the final target boundary frame comprises position coordinates, confidence coefficient and category information of the boundary frame;
(7) And saving the final video.
By comparing various leading edge methods with the pedestrian re-recognition method of the invention, the invention has excellent performance in two data set training as shown in table 1.
TABLE 1 training comparison of this model with other leading edge models
The embodiment of the invention also provides a visual specific pedestrian tracking system based on ReID and Yolov double models, which comprises the following steps:
And a pretreatment module: the method comprises the steps that pedestrian data are obtained by using a plurality of 1280 x 1080 high-definition cameras and SD cameras, and are preprocessed and stored; the pretreatment is specifically as follows: labeling the data, including manual labeling and labeling with a DPM detector; fixing the image size to 256×128; each image is segmented with a probability level of 0.5, each image is decoded into 32-bit anchor point original pixel values, and the RGB channels are normalized. Storing the preprocessed data, wherein naming criteria are as follows: 000N_cNsN_000ABC_0N.jpeg or 000N_cNsN_000ABC_00.jpeg; wherein 000N represents the tag number of each person; cN represents the nth camera; sN represents an nth video clip; 000ABC represents the picture of the 000 th ABC frame of cNsN; 0N represents the Nth detection box on cNsN _000ABC frames.
And a picture extraction module: the method comprises the steps of extracting preprocessed data frame by frame, extracting a person picture in each frame of picture by YOLOv, and storing the picture by extracting a person picture coordinate;
Model training module: the method comprises the steps of obtaining a mark 1501 data set, randomly dividing a test set and a training set, training Resnet models, and obtaining a final model; the method comprises the following steps: the dimension of the Resnet model full-connection layer is modified to N; optimizing the model by adopting an Adam method; selecting identity loss, center loss and enhanced triplet loss as total losses in back propagation; the formula is as follows:
；
Wherein, For identity loss,/>For enhanced triplet loss,/>Is the center loss.
The identity loss formula is as follows:
；
wherein n represents the number of training samples in each batch, provided with a given label Input image/>ThenRepresentation/>Identified as category/>Is used for predicting the probability of (1);
The enhanced triplet loss is as follows:
；
wherein (i, j, k) represents each anchor sample Triplet in each training batch,/>Representing the distance between positive sample pairs,/>Representing the distance between the negative pair of samples; for anchor samples/>And/>Is the corresponding positive sample; /(I)Expressed as anchor sample/>And its corresponding positive sample/>Weight value of Euclidean distance of/>Is a weight value, N represents the positive sample pair number, p represents the environment of the positive sample pair, and N represents the environment of the negative sample pair;
the euclidean distance between two samples is:
；
Wherein, And/>Representation/>And/>Corresponding feature vectors;
the definition formula of k is:
；
the square error is expressed as:
；
The center loss formula is as follows:
；
Wherein, Representing identity/>Is a class center of (c).
And the feature extraction module is used for: the method comprises the steps of inputting a trained model into a picture obtained by a picture extraction module through a transfer learning method, and extracting features;
And an index module: and the method is used for merging the extracted pictures into a tensor, transmitting the tensor to a trained ReID model, extracting and normalizing the characteristics, and finding the index of the minimum average distance by calculating the mean value of Euclidean distances between the query picture and each stored picture.
And an output module: the method comprises the steps of setting corresponding distance thresholds through different scenes, judging the size of the minimum average distance and the distance threshold, and outputting a final target boundary frame when the minimum average distance is smaller than the distance threshold, wherein the final target boundary frame comprises position coordinates, confidence and category information of the boundary frame;
And a storage module: for saving the final video.
Claims (6)
1. A method for visual specific pedestrian tracking based on ReID and Yolov5 double models, comprising the steps of:
(1) Acquiring pedestrian data by using a plurality of 1280 x 1080 high-definition cameras and SD cameras, preprocessing the data and storing the data;
(2) Extracting the preprocessed data frame by frame, extracting a person picture in each frame of picture by YOLOv, and storing the picture by extracting the person picture coordinates;
(3) Acquiring a mark 1501 data set, randomly dividing a test set and a training set, and training Resnet a model to acquire a final model; the method comprises the following steps: the dimension of the Resnet model full-connection layer is modified to N; optimizing the model by adopting an Adam method; selecting identity loss, center loss and enhanced triplet loss as total losses in back propagation; the formula is as follows:
；
Wherein, For identity loss,/>For enhanced triplet loss,/>Is the center loss; the identity loss formula is as follows:
；
wherein n represents the number of training samples in each batch, provided with a given label Input image/>Then/>Representation/>Identified as category/>Is used for predicting the probability of (1);
The enhanced triplet loss is as follows:
；
wherein (i, j, k) represents each anchor sample Triplet in each training batch,/>Representing the distance between positive sample pairs,/>Representing the distance between the negative pair of samples; for anchor samples/>And/>Is the corresponding positive sample; /(I)Expressed as anchor sample/>And its corresponding positive sample/>Weight value of Euclidean distance of/>Is a weight value, N represents the positive sample pair number, p represents the environment of the positive sample pair, and N represents the environment of the negative sample pair;
the euclidean distance between two samples is:
；
Wherein, And/>Representation/>And/>Corresponding feature vectors;
the definition formula of k is:
；
the square error is expressed as:
；
The center loss formula is as follows:
；
Wherein, Representing identity/>Is a class center of (2);
(4) Inputting the trained model into the picture obtained in the step (2) through a transfer learning method, and extracting the characteristics;
(5) Combining the extracted pictures into a tensor, transmitting the tensor to a trained ReID model, extracting and normalizing the characteristics, and finding out the index of the minimum average distance by calculating the mean value of Euclidean distances between the query picture and each stored picture;
(6) Setting corresponding distance thresholds through different scenes, judging the size of the minimum average distance and the distance threshold, and outputting a final target boundary frame when the minimum average distance is smaller than the distance threshold, wherein the final target boundary frame comprises position coordinates, confidence coefficient and category information of the boundary frame;
(7) And saving the final video.
2. The method for visual specific pedestrian tracking based on ReID and Yolov double models according to claim 1, wherein in the step (1), the preprocessing is specifically as follows: labeling the data, including manual labeling and labeling with a DPM detector; fixing the image size to 256×128; each image is segmented with a probability level of 0.5, each image is decoded into 32-bit anchor point original pixel values, and the RGB channels are normalized.
3. The method for visual specific pedestrian tracking based on ReID and Yolov double models according to claim 1, wherein in the step (1), the preprocessed data is stored, and naming criteria are as follows: 000N_cNsN_000ABC_0N.jpeg or 000N_cNsN_000ABC_00.jpeg; wherein 000N represents the tag number of each person; cN represents the nth camera; sN represents an nth video clip; 000ABC represents the picture of the 000 th ABC frame of cNsN; 0N represents the Nth detection box on cNsN _000ABC frames.
4. A visual specific pedestrian tracking system based on ReID and Yolov double models, comprising:
and a pretreatment module: the method comprises the steps that pedestrian data are obtained by using a plurality of 1280 x 1080 high-definition cameras and SD cameras, and are preprocessed and stored;
And a picture extraction module: the method comprises the steps of extracting preprocessed data frame by frame, extracting a person picture in each frame of picture by YOLOv, and storing the picture by extracting a person picture coordinate;
Model training module: the method comprises the steps of obtaining a mark 1501 data set, randomly dividing a test set and a training set, training Resnet models, and obtaining a final model; the method comprises the following steps: the dimension of the Resnet model full-connection layer is modified to N; optimizing the model by adopting an Adam method; selecting identity loss, center loss and enhanced triplet loss as total losses in back propagation; the formula is as follows:
；
Wherein, For identity loss,/>For enhanced triplet loss,/>Is the center loss; the identity loss formula is as follows:
；
wherein n represents the number of training samples in each batch, provided with a given label Input image/>Then/>Representation/>Identified as category/>Is used for predicting the probability of (1);
The enhanced triplet loss is as follows:
；
wherein (i, j, k) represents each anchor sample Triplet in each training batch,/>Representing the distance between positive sample pairs,/>Representing the distance between the negative pair of samples; for anchor samples/>And/>Is the corresponding positive sample; /(I)Expressed as anchor sample/>And its corresponding positive sample/>Weight value of Euclidean distance of/>Is a weight value, N represents the positive sample pair number, p represents the environment of the positive sample pair, and N represents the environment of the negative sample pair;
the euclidean distance between two samples is:
；
Wherein, And/>Representation/>And/>Corresponding feature vectors;
the definition formula of k is:
；
the square error is expressed as:
；
The center loss formula is as follows:
；
Wherein, Representing identity/>Is a class center of (2);
And the feature extraction module is used for: the method comprises the steps of inputting a trained model into a picture obtained by a picture extraction module through a transfer learning method, and extracting features;
And an index module: the method comprises the steps of merging extracted pictures into a tensor, transmitting the tensor to a trained ReID model, extracting and normalizing characteristics, and finding out an index of the minimum average distance by calculating the average value of Euclidean distances between a query picture and each stored picture;
And an output module: the method comprises the steps of setting corresponding distance thresholds through different scenes, judging the size of the minimum average distance and the distance threshold, and outputting a final target boundary frame when the minimum average distance is smaller than the distance threshold, wherein the final target boundary frame comprises position coordinates, confidence and category information of the boundary frame;
And a storage module: for saving the final video.
5. The visual specific pedestrian tracking system based on ReID and Yolov double models according to claim 4, wherein the preprocessing module is configured to: labeling the data, including manual labeling and labeling with a DPM detector; fixing the image size to 256×128; each image is segmented with a probability level of 0.5, each image is decoded into 32-bit anchor point original pixel values, and the RGB channels are normalized.
6. The visual specific pedestrian tracking system based on ReID and Yolov double models of claim 4, wherein the preprocessing module stores the preprocessed data, and naming criteria are as follows: 000N_cNsN_000ABC_0N.jpeg or 000N_cNsN_000ABC_00.jpeg; wherein 000N represents the tag number of each person; cN represents the nth camera; sN represents an nth video clip; 000ABC represents the picture of the 000 th ABC frame of cNsN; 0N represents the Nth detection box on cNsN _000ABC frames.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410161198.5A CN117710903B (en) | 2024-02-05 | 2024-02-05 | Visual specific pedestrian tracking method and system based on ReID and Yolov5 double models |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410161198.5A CN117710903B (en) | 2024-02-05 | 2024-02-05 | Visual specific pedestrian tracking method and system based on ReID and Yolov5 double models |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117710903A CN117710903A (en) | 2024-03-15 |
| CN117710903B true CN117710903B (en) | 2024-05-03 |
Family
ID=90153828
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410161198.5A Active CN117710903B (en) | 2024-02-05 | 2024-02-05 | Visual specific pedestrian tracking method and system based on ReID and Yolov5 double models |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117710903B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111881777A (en) * | 2020-07-08 | 2020-11-03 | 泰康保险集团股份有限公司 | Video processing method and device |
| CN113158891A (en) * | 2021-04-20 | 2021-07-23 | 杭州像素元科技有限公司 | Cross-camera pedestrian re-identification method based on global feature matching |
| CN113408356A (en) * | 2021-05-21 | 2021-09-17 | 深圳市广电信义科技有限公司 | Pedestrian re-identification method, device and equipment based on deep learning and storage medium |
| CN115497056A (en) * | 2022-11-21 | 2022-12-20 | 南京华苏科技有限公司 | Method for detecting lost articles in region based on deep learning |
| CN115841649A (en) * | 2022-11-21 | 2023-03-24 | 哈尔滨工程大学 | Multi-scale people counting method for urban complex scene |
| WO2023093241A1 (en) * | 2021-11-29 | 2023-06-01 | 中兴通讯股份有限公司 | Pedestrian re-identification method and apparatus, and storage medium |
| CN117079309A (en) * | 2023-07-26 | 2023-11-17 | 上海云从企业发展有限公司 | ReID model training method, reID pedestrian recognition method, device and medium |
-
2024
- 2024-02-05 CN CN202410161198.5A patent/CN117710903B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111881777A (en) * | 2020-07-08 | 2020-11-03 | 泰康保险集团股份有限公司 | Video processing method and device |
| CN113158891A (en) * | 2021-04-20 | 2021-07-23 | 杭州像素元科技有限公司 | Cross-camera pedestrian re-identification method based on global feature matching |
| CN113408356A (en) * | 2021-05-21 | 2021-09-17 | 深圳市广电信义科技有限公司 | Pedestrian re-identification method, device and equipment based on deep learning and storage medium |
| WO2023093241A1 (en) * | 2021-11-29 | 2023-06-01 | 中兴通讯股份有限公司 | Pedestrian re-identification method and apparatus, and storage medium |
| CN115497056A (en) * | 2022-11-21 | 2022-12-20 | 南京华苏科技有限公司 | Method for detecting lost articles in region based on deep learning |
| CN115841649A (en) * | 2022-11-21 | 2023-03-24 | 哈尔滨工程大学 | Multi-scale people counting method for urban complex scene |
| CN117079309A (en) * | 2023-07-26 | 2023-11-17 | 上海云从企业发展有限公司 | ReID model training method, reID pedestrian recognition method, device and medium |
Non-Patent Citations (1)
| Title |
|---|
| 基于改进YOLOv5s的视频行人目标检索平台设计与实现;李铭伦;中国优秀硕士学位论文全文数据库 信息科技辑;20230512;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117710903A (en) | 2024-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109344787B (en) | Specific target tracking method based on face recognition and pedestrian re-recognition | |
| US20220415027A1 (en) | Method for re-recognizing object image based on multi-feature information capture and correlation analysis | |
| CN111832514B (en) | Unsupervised pedestrian re-identification method and unsupervised pedestrian re-identification device based on soft multiple labels | |
| CN112004111B (en) | News video information extraction method for global deep learning | |
| CN102549603B (en) | Relevance-based image selection | |
| CN108960184B (en) | Pedestrian re-identification method based on heterogeneous component deep neural network | |
| CN110717411A (en) | Pedestrian re-identification method based on deep layer feature fusion | |
| CN111582178B (en) | Vehicle weight recognition method and system based on multi-azimuth information and multi-branch neural network | |
| CN114067444A (en) | Face spoofing detection method and system based on meta-pseudo label and illumination invariant feature | |
| CN112580657B (en) | Self-learning character recognition method | |
| CN112836675B (en) | Unsupervised pedestrian re-identification method and system for generating pseudo tags based on clusters | |
| CN112215190A (en) | Illegal building detection method based on YOLOV4 model | |
| CN109635647B (en) | Multi-picture multi-face clustering method based on constraint condition | |
| CN112464775A (en) | Video target re-identification method based on multi-branch network | |
| CN114596548A (en) | Target detection method, target detection device, computer equipment and computer-readable storage medium | |
| CN114596546A (en) | Vehicle weight recognition method and device, computer and readable storage medium | |
| CN117710903B (en) | Visual specific pedestrian tracking method and system based on ReID and Yolov5 double models | |
| CN110765940B (en) | Target object statistical method and device | |
| CN112541453A (en) | Luggage weight recognition model training and luggage weight recognition method | |
| CN113435329B (en) | Unsupervised pedestrian re-identification method based on video track feature association learning | |
| CN115937862A (en) | End-to-end container number identification method and system | |
| CN112215189A (en) | Accurate detecting system for illegal building | |
| CN110880022A (en) | Labeling method, labeling device and storage medium | |
| CN116052220B (en) | Pedestrian re-identification method, device, equipment and medium | |
| CN117690031B (en) | SAM model-based small sample learning remote sensing image detection method |
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 |