Pedestrian re-identification method and system for construction site

Abstract

The invention discloses a construction site pedestrian re-identification method. The method comprises the following steps: acquiring an image data set of pedestrian re-identification and preprocessing to obtain a training data set; constructing a construction site pedestrian re-identification initial model based on a ResNet network, and training to obtain a construction site pedestrian re-identification model; and adopting the obtained construction site pedestrian re-identification model to perform pedestrian re-identification on the target construction site. The invention also discloses a system for realizing the construction site pedestrian re-identification method. According to the construction site pedestrian re-identification method and system provided by the invention, feature extraction and comparison in the pedestrian re-identification process of the construction site are realized through the ResNet network, so that the pedestrian re-identification of the construction site can be realized, the reliability is higher, the accuracy is better, and the efficiency is higher.

Description

Technical Field

[0001] The present invention belongs to the field of image data processing, and in particular relates to a method and system for re-identifying pedestrians at a construction site. Background Art

[0002] With the development of economy and technology and the improvement of people's living standards, people pay more and more attention to safety. Construction sites are high-incidence areas for safety accidents, so it is of great significance to carry out refined management of personnel safety issues at construction sites.

[0003] Pedestrian recognition is an important foundation for safety management at construction sites. At present, pedestrian recognition at construction sites still uses manual recognition and punch-in recognition solutions. However, the manual recognition solution is not only time-consuming and labor-intensive, but also inefficient, and the safety of the site safety personnel themselves cannot be guaranteed. The punch-in recognition solution can generally only be used for entry and exit recognition of the construction site area, and cannot perform real-time pedestrian re-identification inside the construction site. Summary of the invention

[0004] One of the purposes of the present invention is to provide a method for re-identifying pedestrians at a construction site with high reliability, good accuracy and high efficiency.

[0005] A second objective of the present invention is to provide a system for implementing the method for re-identifying pedestrians at construction sites.

[0006] The method for re-identifying pedestrians at a construction site provided by the present invention comprises the following steps:

[0007] S1. Obtain an image dataset for person re-identification;

[0008] S2. Preprocess the image data set obtained in step S1 to obtain a training data set;

[0009] S3. Based on the ResNet network, an initial model for pedestrian re-identification at the construction site is constructed;

[0010] S4. Using the training data set obtained in step S2, the initial model for re-identification of construction site pedestrians constructed in step S3 is trained to obtain a model for re-identification of construction site pedestrians;

[0011] S5. Use the construction site pedestrian re-identification model obtained in step S4 to perform pedestrian re-identification on the target construction site.

[0012] The preprocessing described in step S2 specifically includes the following steps:

[0013] Performing data cleaning on the image data set obtained in step S1;

[0014] The images after data cleaning are processed using image enhancement technology to obtain the training data set.

[0015] The step S3 described in which an initial model for pedestrian re-identification at a construction site is constructed based on the ResNet network specifically includes the following steps:

[0016] The ResNet101 network is used as the backbone of the initial model for pedestrian re-identification at construction sites.

[0017] Add the Part branch and use stage 4 of the ResNet101 network as the Global branch;

[0018] Use the output of stage 3 of the ResNet101 network as the input of the Part branch and the Global branch;

[0019] The Part branch includes four residual modules Player4.1, Player4.2, Player4.3 and Player4.4 connected in series. The fourth residual module performs a pooling operation with stride=1 to improve the fine-grained features.

[0020] The Global branch includes four residual modules Glayer4.1, Glayer4.2, Glayer4.3 and Glayer4.4 connected in series;

[0021] The output of stage3 of the ResNet101 network is concatenated with the output of the residual module Player4.1 in the Part branch, and then processed by global maximum pooling to obtain the first feature layer4.1_p;

[0022] The output of stage3 of the ResNet101 network is concatenated with the output of the residual module Glayer4.1 in the Global branch, and then processed by global maximum pooling to obtain the second feature layer4.1_g;

[0023] After the output of the Part branch is processed by global average pooling, the third feature layer4.4_part is obtained;

[0024] The output of the Global branch is processed by global average pooling to obtain the fourth feature layer4.4_global;

[0025] After concatenating the first feature layer4.1_p, the second feature layer4.1_g, the third feature layer4.4_part and the fourth feature layer4.4_global, the dimensionality is reduced through a 1*1 convolution layer to obtain the final features for person re-identification.

[0026] The obtained features of pedestrian re-identification are used to measure the similarity with the features of pedestrians in the previously stored pedestrian database: if the similarity measurement result is greater than the set threshold, it means that the comparison is successful and the result of pedestrian re-identification is obtained, otherwise it means that the comparison fails.

[0027] The training described in step S4 specifically includes the following steps:

[0028] During training, the cross entropy loss function is used to train the constructed initial model for pedestrian re-identification at construction sites.

[0029] The present invention also provides a system for implementing the method for pedestrian re-identification at a construction site, comprising a data acquisition module, a data processing module, a model building module, a model training module and a pedestrian re-identification module; the data acquisition module, the data processing module, the model building module, the model training module and the pedestrian re-identification module are connected in series in sequence; the data acquisition module is used to acquire an image data set for pedestrian re-identification and upload the data information to the data processing module; the data processing module is used to pre-process the acquired image data set according to the received data information to obtain a training data set, and upload the data information to the model building module; the model building module is used to construct an initial model for pedestrian re-identification at a construction site based on a ResNet network according to the received data information, and upload the data information to the model training module; the model training module is used to train the constructed initial model for pedestrian re-identification at a construction site using the obtained training data set according to the received data information, obtain a pedestrian re-identification model for a construction site, and upload the data information to the pedestrian re-identification module; the pedestrian re-identification module is used to re-identify pedestrians at a target site according to the received data information using the obtained construction site pedestrian re-identification model.

[0030] The method and system for pedestrian re-identification at construction sites provided by the present invention realize feature extraction and comparison in the process of pedestrian re-identification at construction sites through a ResNet network. Therefore, the present invention can not only realize pedestrian re-identification at construction sites, but also has higher reliability, better accuracy and higher efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The figure is a schematic diagram of the method flow of the present invention.

[0032] Figure 2 Schematic diagram of the functional modules of the system of the present invention. DETAILED DESCRIPTION

[0033] like Figure 1 The method flow chart of the method of the present invention is shown as follows: The method for re-identifying pedestrians at a construction site disclosed by the present invention comprises the following steps:

[0034] S1. Obtain an image dataset for person re-identification;

[0035] S2. Preprocess the image data set obtained in step S1 to obtain a training data set; specifically, the steps include:

[0036] Performing data cleaning on the image data set obtained in step S1;

[0037] The images after data cleaning are processed using image enhancement technology to obtain a training data set;

[0038] S3. Based on the ResNet network, an initial model for pedestrian re-identification at the construction site is constructed; the specific steps include:

[0039] The ResNet101 network is used as the backbone of the initial model for pedestrian re-identification at construction sites.

[0040] Add the Part branch and use stage 4 of the ResNet101 network as the Global branch;

[0041] Use the output of stage 3 of the ResNet101 network as the input of the Part branch and the Global branch;

[0042] The Part branch includes four residual modules Player4.1, Player4.2, Player4.3 and Player4.4 connected in series. The fourth residual module performs a pooling operation with stride=1 to improve fine-grained features and retain more tiny features of pedestrian images.

[0043] The Global branch includes four residual modules Glayer4.1, Glayer4.2, Glayer4.3 and Glayer4.4 connected in series;

[0044] After splicing the output of stage3 of the ResNet101 network with the output of the residual module Player4.1 in the Part branch, the first feature layer4.1_p is obtained through global maximum pooling. Global maximum pooling can eliminate the influence of background clutter and make the network pay more attention to the limb features and real-time features of the target pedestrian.

[0045] The output of stage3 of the ResNet101 network is concatenated with the output of the residual module Glayer4.1 in the Global branch, and then processed by global maximum pooling to obtain the second feature layer4.1_g. Global maximum pooling can eliminate the influence of background clutter and make the network pay more attention to the limb features and real-time features of the target pedestrian.

[0046] The output of the Part branch is processed by global average pooling to obtain the third feature layer4.4_part; global average pooling can make full use of the network to extract features;

[0047] The output of the Global branch is processed by global average pooling to obtain the fourth feature layer4.4_global; global average pooling can make full use of the network to extract features;

[0048] After concatenating the first feature layer4.1_p, the second feature layer4.1_g, the third feature layer4.4_part and the fourth feature layer4.4_global, the dimensionality is reduced through a 1*1 convolution layer to obtain the final pedestrian re-identification feature; this feature contains both low-level features and high-level semantics, has strong robustness, and can improve the accuracy of pedestrian re-identification; through the dimensionality reduction operation, a balance between accuracy and efficiency is achieved;

[0049] The obtained pedestrian re-identification features are measured for similarity with the features of pedestrians in the previously stored pedestrian database: if the similarity measurement result is greater than the set threshold, it means that the comparison is successful and the pedestrian re-identification result is obtained, otherwise it means that the comparison fails;

[0050] S4. Using the training data set obtained in step S2, the initial model for re-identification of construction site pedestrians constructed in step S3 is trained to obtain a model for re-identification of construction site pedestrians;

[0051] During training, the cross entropy loss function is used to train the initial model of pedestrian re-identification at the construction site.

[0052] S5. Use the construction site pedestrian re-identification model obtained in step S4 to re-identify pedestrians at the target construction site. In specific implementation, the trained construction site pedestrian re-identification model can be deployed to the monitoring room of the construction site, and the pedestrian information of the construction site can be obtained through various cameras. Then, the pedestrian re-identification can be realized through the host computer of the monitoring room.

[0053] like Figure 2The functional module schematic diagram of the system of the present invention is shown as follows: the system disclosed in the present invention for realizing the method for re-identifying pedestrians on construction sites comprises a data acquisition module, a data processing module, a model building module, a model training module and a pedestrian re-identification module; the data acquisition module, the data processing module, the model building module, the model training module and the pedestrian re-identification module are connected in series in sequence; the data acquisition module is used to acquire an image data set for pedestrian re-identification and upload the data information to the data processing module; the data processing module is used to pre-process the acquired image data set according to the received data information to obtain a training data set, and upload the data information to the model building module; the model building module is used to construct an initial model for re-identifying pedestrians on construction sites based on a ResNet network according to the received data information, and upload the data information to the model training module; the model training module is used to train the constructed initial model for re-identifying pedestrians on construction sites according to the received data information using the obtained training data set to obtain a pedestrian re-identification model for construction sites, and upload the data information to the pedestrian re-identification module; the pedestrian re-identification module is used to re-identify pedestrians on a target site according to the received data information using the obtained pedestrian re-identification model for construction sites.

Claims

1. A method for re-identifying pedestrians at a construction site, comprising the following steps: S1. Obtain an image dataset for person re-identification; S2. Preprocess the image data set obtained in step S1 to obtain a training data set; S3. Based on the ResNet network, an initial model for pedestrian re-identification at the construction site is constructed; S4. Using the training data set obtained in step S2, the initial model for re-identification of construction site pedestrians constructed in step S3 is trained to obtain a model for re-identification of construction site pedestrians; S5. Use the construction site pedestrian re-identification model obtained in step S4 to perform pedestrian re-identification on the target construction site.

2. The method for re-identifying pedestrians at a construction site according to claim 1, characterized in that The preprocessing described in step S2 specifically includes the following steps: Performing data cleaning on the image data set obtained in step S1; The images after data cleaning are processed using image enhancement technology to obtain the training data set.

3. The method for re-identifying pedestrians at a construction site according to claim 2, characterized in that The step S3 described in which an initial model for pedestrian re-identification at a construction site is constructed based on the ResNet network specifically includes the following steps: The ResNet101 network is used as the backbone of the initial model for pedestrian re-identification at construction sites. Add the Part branch and use stage 4 of the ResNet101 network as the Global branch; Use the output of stage 3 of the ResNet101 network as the input of the Part branch and the Global branch; The Part branch includes four residual modules Player4.1, Player4.2, Player4.3 and Player4.4 connected in series. The fourth residual module performs a pooling operation with stride=1 to improve the fine-grained features. The Global branch includes four residual modules Glayer4.1, Glayer4.2, Glayer4.3 and Glayer4.4 connected in series; The output of stage3 of the ResNet101 network is concatenated with the output of the residual module Player4.1 in the Part branch, and then processed by global maximum pooling to obtain the first feature layer4.1_p; The output of stage3 of the ResNet101 network is concatenated with the output of the residual module Glayer4.1 in the Global branch, and then processed by global maximum pooling to obtain the second feature layer4.1_g; After the output of the Part branch is processed by global average pooling, the third feature layer4.4_part is obtained; The output of the Global branch is processed by global average pooling to obtain the fourth feature layer4.4_global; After concatenating the first feature layer4.1_p, the second feature layer4.1_g, the third feature layer4.4_part and the fourth feature layer4.4_global, the dimensionality is reduced through a 1*1 convolution layer to obtain the final features for person re-identification. The obtained features of pedestrian re-identification are used to measure the similarity with the features of pedestrians in the previously stored pedestrian database: if the similarity measurement result is greater than the set threshold, it means that the comparison is successful and the result of pedestrian re-identification is obtained, otherwise it means that the comparison fails.

4. The method for re-identifying pedestrians at a construction site according to claim 3, characterized in that The training described in step S4 specifically includes the following steps: During training, the cross entropy loss function is used to train the constructed initial model for pedestrian re-identification at construction sites.

5. A system for implementing the method for re-identifying pedestrians at a construction site according to any one of claims 1 to 4, characterized in that It includes a data acquisition module, a data processing module, a model building module, a model training module and a pedestrian re-identification module; the data acquisition module, the data processing module, the model building module, the model training module and the pedestrian re-identification module are connected in series in sequence; the data acquisition module is used to obtain an image data set for pedestrian re-identification and upload the data information to the data processing module; The data processing module is used to pre-process the acquired image data set according to the received data information, obtain the training data set, and upload the data information to the model construction module; The model building module is used to build an initial model for pedestrian re-identification at the construction site based on the received data information and the ResNet network, and upload the data information to the model training module; The model training module is used to train the constructed initial model of pedestrian re-identification at the construction site based on the received data information and the obtained training data set, obtain the pedestrian re-identification model at the construction site, and upload the data information to the pedestrian re-identification module; the pedestrian re-identification module is used to perform pedestrian re-identification on the target construction site based on the received data information and the obtained pedestrian re-identification model at the construction site.

Images