Pedestrian flow statistical method and system based on human head and shoulder ReID technology combination

Abstract

The invention discloses a human body head and shoulder ReID technology combination-based people flow statistics method and system, and belongs to the technical field of people flow statistics. According to the method, the structural features of the personnel in the video are obtained by adopting a head and shoulder recognition method and a personnel feature segmentation extraction model, the accuracy of extracting the structural features of the personnel in a crowded or shielded actual scene is improved, meanwhile, the accuracy of cross-camera personnel recognition is improved through similarity comparison of the structural features of the personnel, and the accuracy of cross-camera personnel recognition is improved. And continuous tracking of people is facilitated, and the accuracy of people flow statistics is improved.

Description

Technical Field

[0001] The present application relates to the technical field of human flow counting, and in particular to a human flow counting method and system based on the combination of human head and shoulder ReID technology. Background Art

[0002] With the rapid development of smart cities, crowd flow monitoring and real-time counting in public places have become increasingly important. This is not only a basic requirement for urban management and planning, but also a key link in improving public safety, resource allocation efficiency and optimizing citizen experience. Traditional video-based crowd counting methods have played a certain role in this regard. These methods include background subtraction, target detection and tracking, virtual counting lines, and optical flow methods.

[0003] Background subtraction technology is usually used to extract foreground objects, i.e., individuals that are moving, from videos. Common background subtraction methods, such as frame difference method and mixed Gaussian model, perform relatively well in static environments. However, in complex scenes, these methods may have large errors due to factors such as lighting changes and dynamic backgrounds. For example, sudden changes in light or moving objects in the background (such as leaves swaying in the wind) may be mistakenly identified as foreground objects, thus affecting the accuracy of statistics.

[0004] Target detection and tracking technology achieves crowd counting by identifying and tracking moving targets in the video. Traditional methods include contour detection, color segmentation, mean shift, etc. These methods are more effective when the target is single and the environment is simple. However, in actual scenes with dense crowds or occlusions, these algorithms often fail due to mutual occlusion of individuals or changes in the target's shape, resulting in frequent false detection and missed detection problems.

[0005] The virtual counting line and area method sets some virtual lines or areas of interest and counts the number of times the target passes through these locations. This method is practical in areas with one-way flow and clear channels. However, its accuracy is significantly reduced in open areas with multi-directional flow and unfixed paths, especially in high-density crowds, where overlapping and intersecting paths will make the counting line method ineffective.

[0006] The optical flow method analyzes the overall motion trend by capturing the speed and direction of the pixels in the video frame. This has advantages in the detection of fast-moving individuals. However, the high computational complexity and sensitivity to noise of the optical flow method limit its use in real-time applications.

[0007] In summary, although these traditional methods are still valuable in certain specific application environments, they are easily affected by factors such as lighting changes and dynamic backgrounds in complex environments, resulting in limited accuracy. In particular, in crowded and severely occluded environments, false detections and missed detections are prone to occur. Summary of the invention

[0008] The present application aims to provide a method and system for counting human traffic based on the combination of human head and shoulder ReID (Re-Identification) technology to solve the deficiencies in the prior art. The technical problem to be solved by the present application is achieved through the following technical solutions.

[0009] On the one hand, a method for counting human traffic based on human head and shoulder ReID technology is provided, the method comprising:

[0010] S1. Collect video information through the camera, use the head and shoulder recognition method and the personnel feature segmentation and extraction model to obtain the structural features of the personnel in the video;

[0011] S2, storing the structured features of the personnel in the video in a database to obtain a personnel structured feature database;

[0012] S3, arranging multiple cameras at different positions, respectively collecting and processing the personnel information in the video, and obtaining the personnel structured features of the personnel appearing in each camera;

[0013] S4, performing a similarity comparison on the personnel structured features of the personnel appearing in each camera respectively, if the comparison requirements are met, the personnel structured features are fused, if the comparison requirements are not met, the personnel structured features are not fused, and the fused personnel structured features are obtained;

[0014] S5. Compare the fused personnel structured features with the personnel structured feature database. If they meet the requirements of the personnel structured feature database, update the personnel data table. If they do not meet the requirements of the personnel structured feature database, create a new personnel data table to obtain the personnel data table of the personnel appearing in the camera.

[0015] S6. According to the personnel data table of the personnel appearing in the camera, the personnel trajectory of the personnel appearing in the camera is obtained through path analysis;

[0016] S7. According to the trajectory of the people appearing in the camera, statistical information of human flow is obtained through statistical analysis.

[0017] In an optional implementation, in S1, video information is collected by a camera, and a head and shoulder recognition method and a personnel feature segmentation and extraction model are used to obtain structured features of personnel in the video, including:

[0018] S11, according to the video information collected by the camera, marking the person in the video information to obtain the person frame;

[0019] S12, obtaining a head-shoulder frame based on the person frame by using a head-shoulder detection model and ReID technology;

[0020] S13, according to the head-shoulder frame, using the head-shoulder tracking module, obtain the head-shoulder ID (Identity document, exclusive identification number);

[0021] S14, according to the video information collected by the camera, a personnel feature vector is obtained through a personnel feature segmentation extraction model;

[0022] S15. Obtain structured features of the person in the video according to the head and shoulder ID and the person feature vector.

[0023] In an optional implementation, the personnel feature segmentation and extraction model is a model pre-trained using a ResNet50 (Deep Residual Network) network with a prediction layer removed and a private personnel dataset without background.

[0024] In an optional implementation, in S14, according to the video information collected by the camera, a personnel feature vector is obtained through a personnel feature segmentation extraction model, including:

[0025] S141, based on the video information collected by the camera, the person can be segmented to exclude the background interference in the person frame, and a human mask (Human Mask) can be obtained;

[0026] S142. According to the human body mask, obtain a person feature vector by extracting obvious features.

[0027] In an optional implementation, the method for segmenting a person to exclude background interference in the person frame includes: random cropping, rotation and color change.

[0028] In an optional implementation, in S4, the personnel structured features of the personnel appearing in each camera are respectively compared for similarity. If the comparison requirements are met, the personnel structured features are fused. If the comparison requirements are not met, the personnel structured features are not fused, and the fused personnel structured features are obtained, including:

[0029] S41, predicting the structural features of the personnel appearing in each camera at the next moment in combination with the dynamic model and measurement data of the system;

[0030] S42, performing a similarity comparison between the personnel structured features of the personnel appearing in each camera at the next moment and the personnel structured features of the personnel appearing in each camera;

[0031] S43. If the comparison requirements are met, the personnel structured features are fused; if the comparison requirements are not met, the personnel structured features are not fused, and the fused personnel structured features are obtained.

[0032] In an optional embodiment, the training step of the dynamic model of the system includes:

[0033] S411. Obtaining a state estimation value at a previous moment and a state estimation value at a current moment according to the structural characteristics of the personnel;

[0034] S412, predicting the current state estimation value based on the state estimation value at the previous moment, and obtaining the current state prediction value;

[0035] S413, obtaining a Kalman gain by difference calculation based on the state estimation value at the previous moment and the state estimation value at the current moment;

[0036] S414. Adjust the current state prediction value according to the Kalman gain so that the current state prediction value is consistent with the current state estimation value.

[0037] In an optional implementation manner, the method for comparing the fused personnel structured features with the personnel structured feature database includes: Euclidean distance comparison method.

[0038] In an optional implementation, the statistical analysis method includes: time series analysis, cluster analysis and association rule mining.

[0039] On the other hand, a human flow statistics system based on the combination of human head and shoulder ReID technology is provided, which can be used for a human flow statistics method based on the combination of human head and shoulder ReID technology, the system comprising: a first feature extraction module, a database establishment module, a second feature extraction module, a fusion processing module, a data table acquisition module, a trajectory acquisition module and a human flow statistics module;

[0040] The first feature extraction module is used to collect video information through a camera, and use a head and shoulder recognition method and a personnel feature segmentation extraction model to obtain structured features of personnel in the video;

[0041] The database establishment module is used to store the structured features of people in the video in the database to obtain a database of structured features of people;

[0042] The second feature extraction module is used to arrange multiple cameras at different positions, respectively collect and process the personnel information in the video, and obtain the personnel structured features of the personnel appearing in each camera;

[0043] The fusion processing module is used to compare the similarity of the personnel structured features of the personnel appearing in each camera respectively. If the comparison requirements are met, the personnel structured features are fused. If the comparison requirements are not met, the personnel structured features are not fused to obtain the fused personnel structured features.

[0044] The data table acquisition module is used to compare the fused personnel structured features with the personnel structured feature database. If the fused personnel structured features meet the requirements of the personnel structured feature database, the personnel data table is updated. If the fused personnel structured features do not meet the requirements of the personnel structured feature database, a new personnel data table is created to obtain the personnel data table of the personnel appearing in the camera.

[0045] The trajectory acquisition module is used to obtain the trajectory of the person appearing in the camera through path analysis according to the personnel data table of the person appearing in the camera;

[0046] The human flow statistics module is used to obtain human flow statistics information through statistical analysis of the human trajectories that appear in the camera.

[0047] The embodiments of the present application include the following advantages:

[0048] The embodiments of the present application provide a method and system for counting people flow based on the combination of human head and shoulder ReID technology. On the one hand, the application adopts a head and shoulder recognition method and a personnel feature segmentation and extraction model to obtain the structured features of people in the video, thereby improving the accuracy of extracting the structured features of people in actual scenes with dense crowds or occlusions. On the other hand, by comparing the similarity of personnel structured features, the accuracy of personnel identification across cameras is improved, which helps to continuously track personnel and improve the accuracy of people flow statistics. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flowchart of the steps of an embodiment of a human flow counting method based on the human head and shoulder ReID technology of the present application;

[0050] Figure 2 This is a flowchart of the steps of obtaining structural features of people in a video in an embodiment of a human flow statistics method based on the human head and shoulder ReID technology combined with the present application;

[0051] Figure 3 This is a flowchart of the steps of obtaining a person feature vector in an embodiment of a method for counting human traffic based on the combination of human head and shoulder ReID technology of the present application;

[0052] Figure 4 This is a flowchart of the steps of obtaining the structural features of people after fusion processing in the embodiment of the human flow statistics method based on the human head and shoulder ReID technology combined with the present application;

[0053] Figure 5 It is a flow chart of the training steps of the dynamic model of the system in the embodiment of the human flow statistics method based on the human head and shoulder ReID technology combined with the present application;

[0054] Figure 6 This is a system block diagram of an embodiment of a human flow statistics method based on the combination of human head and shoulder ReID technology of the present application. DETAILED DESCRIPTION

[0055] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0056] This embodiment is a scene of counting the flow of people in a community, such as Figure 1 The flowchart of the steps of the embodiment of the method for counting human traffic based on the combination of human head and shoulder ReID (Re-Identification) technology of the present application is shown. The present application provides a method for counting human traffic based on the combination of human head and shoulder ReID technology, and the method includes:

[0057] S1. Collect video information through the camera, use the head and shoulder recognition method and the personnel feature segmentation and extraction model to obtain the structural features of the personnel in the video;

[0058] Specifically, Figure 2 The flowchart shown is a step flow chart of the steps for obtaining structured features of people in the video in the embodiment of the crowd statistics method based on the human head and shoulder ReID technology combined with the present application. Video information is collected according to the cameras in the community, and the people in the video information in the community are marked to obtain a person frame. Then, based on the person frame, a head and shoulder detection model and ReID technology are used to obtain a head and shoulder frame. ReID technology mainly relies on deep learning methods, especially convolutional neural networks, to extract feature representations of objects. These features should be highly distinguishable so that the same object can be identified in different scenarios.

[0059] The ReID technology usually includes the following aspects:

[0060] Data collection and preprocessing: Collect a large amount of image or video data, which should include different individuals captured by multiple cameras;

[0061] Feature extraction: Use pre-trained convolutional neural networks (such as ResNet, VGG, etc.) as the basic model to extract features from images;

[0062] Feature matching, by calculating the similarity between feature vectors extracted from different images (such as Euclidean distance, cosine similarity, etc.), to determine whether they belong to the same object. A variety of strategies can be used to improve the accuracy of matching, such as introducing attention mechanisms, multi-scale feature fusion, etc.;

[0063] Loss function and optimization,During the training process, the network parameters are adjusted through the back-propagation algorithm,,the loss function is minimized to improve the recognition ability of the,model.

[0064] Then, according to the head-shoulder frame, the head-shoulder ID (Identity document, exclusive identification number) is obtained by using the head-shoulder tracking module;

[0065] At the same time, based on the video information collected by the camera, color transformation is used to eliminate background interference in the person frame, and the ResNet50 (Deep Residual Network) network without the prediction layer and the model pre-trained with the private person data set without background are used to segment the people and obtain the human mask (Human Mask). Based on the human mask, the person feature vector is obtained by extracting obvious features.

[0066] Finally, according to the head and shoulder ID and the person feature vector, the structured features of the person in the video are obtained.

[0067] S2, storing the structured features of the personnel in the video in a database to obtain a personnel structured feature database;

[0068] S3, arranging multiple cameras at different positions, respectively collecting and processing the personnel information in the video, and obtaining the personnel structured features of the personnel appearing in each camera;

[0069] S4, performing a similarity comparison on the personnel structured features of the personnel appearing in each camera respectively, if the comparison requirements are met, the personnel structured features are fused, if the comparison requirements are not met, the personnel structured features are not fused, and the fused personnel structured features are obtained;

[0070] Specifically, Figure 4 The flowchart of the steps of obtaining the structured features of people after fusion processing in the embodiment of the human flow statistics method based on the human head and shoulder ReID technology of the present application is shown. First, according to the structured features of people appearing in each camera, combined with the dynamic model and measurement data of the system, the structured features of people appearing in each camera at the next moment are predicted; wherein, Figure 5The flowchart of the training steps of the dynamic model of the system in the embodiment of the human flow statistics method based on the human head and shoulder ReID technology of the present application is shown. The training steps of the dynamic model of the system include:

[0071] S411. Obtaining a state estimation value at a previous moment and a state estimation value at a current moment according to the structural characteristics of the personnel;

[0072] S412, predicting the current state estimation value based on the state estimation value at the previous moment, and obtaining the current state prediction value;

[0073] S413, obtaining a Kalman gain by difference calculation based on the state estimation value at the previous moment and the state estimation value at the current moment;

[0074] S414. Adjust the current state prediction value according to the Kalman gain so that the current state prediction value is consistent with the current state estimation value.

[0075] Then, the similarity between the personnel structured features of the personnel appearing in each camera at the next moment and the personnel structured features of the personnel appearing in each camera is compared.

[0076] Finally, if the comparison requirements are met, the personnel structured features are fused; if the comparison requirements are not met, the personnel structured features are not fused, and the fused personnel structured features are obtained.

[0077] S5. Compare the fused personnel structured features with the personnel structured feature database by using the Euclidean distance comparison method. If the fused personnel structured features meet the requirements of the personnel structured feature database, update the personnel data table. If the fused personnel structured features do not meet the requirements of the personnel structured feature database, create a new personnel data table to obtain the personnel data table of the personnel appearing in the camera.

[0078] S6. According to the personnel data table of the personnel appearing in the camera, the personnel trajectory of the personnel appearing in the camera is obtained through path analysis;

[0079] S7. According to the trajectory of the people appearing in the camera, statistical information of the flow of people in the community is obtained through statistical analysis.

[0080] Specifically, the statistical analysis method includes: time series analysis method, cluster analysis method and association rule mining method.

[0081] Figure 6A human flow counting system based on the combination of human head and shoulder ReID technology is shown according to an exemplary embodiment. The system can be used for a human flow counting method based on the combination of human head and shoulder ReID technology. The system includes: a first feature extraction module, a database establishment module, a second feature extraction module, a fusion processing module, a data table acquisition module, a trajectory acquisition module and a human flow counting module;

[0082] The first feature extraction module is used to collect video information through a camera, and use a head and shoulder recognition method and a personnel feature segmentation extraction model to obtain structured features of personnel in the video;

[0083] The database establishment module is used to store the structured features of people in the video in the database to obtain a database of structured features of people;

[0084] The second feature extraction module is used to arrange multiple cameras at different positions, respectively collect and process the personnel information in the video, and obtain the personnel structured features of the personnel appearing in each camera;

[0085] The fusion processing module is used to compare the similarity of the personnel structured features of the personnel appearing in each camera respectively. If the comparison requirements are met, the personnel structured features are fused. If the comparison requirements are not met, the personnel structured features are not fused to obtain the fused personnel structured features.

[0086] The data table acquisition module is used to compare the fused personnel structured features with the personnel structured feature database. If the fused personnel structured features meet the requirements of the personnel structured feature database, the personnel data table is updated. If the fused personnel structured features do not meet the requirements of the personnel structured feature database, a new personnel data table is created to obtain the personnel data table of the personnel appearing in the camera.

[0087] The trajectory acquisition module is used to obtain the trajectory of the person appearing in the camera through path analysis according to the personnel data table of the person appearing in the camera;

[0088] The human flow statistics module is used to obtain human flow statistics information through statistical analysis of the human trajectories that appear in the camera.

[0089] The embodiment of the present application provides a method and system for counting people flow based on the combination of human head and shoulder ReID technology. The application adopts a head and shoulder recognition method and a personnel feature segmentation and extraction model to obtain the structured features of people in the video, thereby improving the accuracy of extracting the structured features of people in actual scenes with dense crowds or occlusions. At the same time, by comparing the similarity of personnel structured features, the accuracy of personnel identification across cameras is improved, which helps to continuously track personnel and improve the accuracy of people flow statistics.

[0090] It should be noted that the above detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present application belongs.

[0091] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0092] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0093] In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products, or apparatuses.

[0094] In the above detailed description, reference is made to the accompanying drawings, which form a part of this document. In the accompanying drawings, similar symbols typically identify similar components unless the context indicates otherwise. The illustrated embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be used, and other changes may be made, without departing from the spirit or scope of the subject matter presented herein.

[0095] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for counting human traffic based on human head and shoulder ReID technology, characterized in that: The method comprises: S1. Collect video information through the camera, use the head and shoulder recognition method and the personnel feature segmentation and extraction model to obtain the structural features of the personnel in the video; S2, storing the structured features of the personnel in the video in a database to obtain a structured features database of personnel; S3, arranging multiple cameras at different positions, respectively collecting and processing the personnel information in the video, and obtaining the personnel structured features of the personnel appearing in each camera; S4, respectively performing a similarity comparison on the personnel structured features of the personnel appearing in each camera, and if the similarity meets the comparison requirements, the personnel structured features are fused, and if the similarity does not meet the comparison requirements, the personnel structured features are not fused, and the fused personnel structured features are obtained; S5, comparing the fused personnel structured features with the personnel structured feature database, if they meet the requirements of the personnel structured feature database, updating the personnel data table, if they do not meet the requirements of the personnel structured feature database, creating a new personnel data table to obtain the personnel data table of the personnel appearing in the camera; S6. According to the personnel data table of the personnel appearing in the camera, the personnel trajectory of the personnel appearing in the camera is obtained through path analysis; S7. Obtaining human flow statistics information through statistical analysis based on the trajectory of the people appearing in the camera.

2. The method for counting human traffic based on the combination of human head and shoulder ReID technology according to claim 1 is characterized in that: In S1, video information is collected by a camera, and the head and shoulder recognition method and the personnel feature segmentation and extraction model are used to obtain the structured features of the personnel in the video, including: S11, according to the video information collected by the camera, marking the person in the video information to obtain the person frame; S12, obtaining a head-shoulder frame according to the person frame by using a head-shoulder detection model and ReID technology; S13, obtaining a head-shoulder ID using a head-shoulder tracking module according to the head-shoulder frame; S14, according to the video information collected by the camera, a personnel feature vector is obtained through a personnel feature segmentation extraction model; S15. Obtain structured features of the person in the video according to the head and shoulder ID and the person feature vector.

3. The method for counting human traffic based on the combination of human head and shoulder ReID technology according to claim 2 is characterized in that: The personnel feature segmentation and extraction model is a model pre-trained using a ResNet50 network with the prediction layer removed and a private personnel dataset without background.

4. The method for counting human traffic based on the combination of human head and shoulder ReID technology according to claim 2 is characterized in that: In S14, according to the video information collected by the camera, a personnel feature vector is obtained through a personnel feature segmentation extraction model, including: S141, based on the video information collected by the camera, the person can be segmented to exclude the background interference in the person frame, and a human body mask can be obtained; S142. Obtain a person feature vector by extracting obvious features according to the human body mask.

5. The method for counting human traffic based on the combination of human head and shoulder ReID technology according to claim 4 is characterized in that: The method for segmenting a person and excluding background interference in the person frame includes: random cropping, rotation and color change.

6. The method for counting human traffic based on the combination of human head and shoulder ReID technology according to claim 1 is characterized in that: In S4, the personnel structured features of the personnel appearing in each camera are respectively compared for similarity. If the comparison requirements are met, the personnel structured features are fused. If the comparison requirements are not met, the personnel structured features are not fused, and the fused personnel structured features are obtained, including: S41, predicting the personnel structured features of the personnel appearing in each camera at the next moment in combination with the dynamic model and measurement data of the system according to the personnel structured features of the personnel appearing in each camera; S42, performing a similarity comparison between the personnel structured features of the personnel appearing in each camera at the next moment and the personnel structured features of the personnel appearing in each camera; S43. If the comparison requirements are met, the personnel structured features are fused; if the comparison requirements are not met, the personnel structured features are not fused, and the fused personnel structured features are obtained.

7. The method for counting human traffic based on the combination of human head and shoulder ReID technology according to claim 6 is characterized in that: The training steps of the dynamic model of the system include: S411. Obtaining a state estimation value at a previous moment and a state estimation value at a current moment according to the structural characteristics of the personnel; S412, predicting the current state estimation value according to the state estimation value at the previous moment to obtain the current state prediction value; S413, obtaining a Kalman gain by difference calculation according to the state estimation value at the previous moment and the state estimation value at the current moment; S414. Adjust the state prediction value at this moment according to the Kalman gain so that the state prediction value at this moment is consistent with the state estimation value at this moment.

8. The method for counting human traffic based on the combination of human head and shoulder ReID technology according to claim 1 is characterized in that: The method for comparing the fused personnel structured features with the personnel structured feature database includes: a Euclidean distance comparison method.

9. The method for counting human traffic based on the combination of human head and shoulder ReID technology according to claim 1 is characterized in that: The statistical analysis method includes: time series analysis method, cluster analysis method and association rule mining method.

10. A human flow counting system based on human head and shoulder ReID technology, characterized in that: The system comprises: The first feature extraction module is used to collect video information through a camera, and use a head and shoulder recognition method and a personnel feature segmentation extraction model to obtain structured features of personnel in the video; A database establishment module, used for storing the structured features of personnel in the video in a database to obtain a structured features database of personnel; The second feature extraction module is used to arrange multiple cameras at different positions, respectively collect and process the personnel information in the video, and obtain the personnel structured features of the personnel appearing in each camera; A fusion processing module is used to compare the similarity of the personnel structured features of the personnel appearing in each camera respectively, and if the comparison requirements are met, the personnel structured features are fused; if the comparison requirements are not met, the personnel structured features are not fused, so as to obtain the personnel structured features after fusion processing; A data table acquisition module is used to compare the fused personnel structured features with the personnel structured feature database, and if they meet the requirements of the personnel structured feature database, update the personnel data table; if they do not meet the requirements of the personnel structured feature database, create a new personnel data table to obtain the personnel data table of the personnel appearing in the camera; A trajectory acquisition module, used to obtain the trajectory of the person appearing in the camera through path analysis according to the personnel data table of the person appearing in the camera; The human flow statistics module is used to obtain human flow statistics information by statistical analysis of the human trajectories of the people appearing in the camera.

Images