CN115169673A - Intelligent campus epidemic risk monitoring and early warning system and method - Google Patents

Abstract

The invention discloses a smart campus epidemic situation risk monitoring and early warning system and a smart campus epidemic situation risk monitoring and early warning method, wherein the system comprises the following steps: including image data acquisition processing module, the identity differentiates the module, the target detection module, track identification module and epidemic situation risk detection module, image data acquisition processing module is used for gathering campus crowd flow condition, carry out the preliminary treatment according to flow data, including data enhancement, preliminary treatment such as data mark and data set division, the identity is differentiateed the module and is used for discerning target identity information based on pedestrian face identification and pedestrian gesture recognition, target detection module is used for carrying out pedestrian target state detection to image data, the track identification module is used for detecting pedestrian's action scope and action mode, epidemic situation risk detection module is used for differentiateing campus flow, study and judge and the early warning to campus crowd flow condition. The campus epidemic situation control system can provide refined track information of people going out in a campus, and realize accurate campus epidemic situation prevention and control by assisting campus epidemic situation flow-regulation data.

Description

A smart campus epidemic risk monitoring and early warning system and method

technical field

The invention belongs to the technical field of data monitoring, and in particular relates to a smart campus epidemic risk monitoring and early warning system and method.

Background technique

The outbreak of the epidemic has brought a major impact on the country's economy and society. As a crowded place, campus is a key link and important area for epidemic prevention and control. Compared with crowds in other public places, campus crowds have the typical characteristics of strong sociality and interaction, concentrated activity places, and high overlap of flow trajectories, making it difficult for existing epidemic prevention and control technology products to play a key role in campus epidemic prevention and control.

For the risk monitoring and prevention and control of the new crown epidemic, scholars at home and abroad have carried out a series of technical explorations, and have achieved rich research results in some fields, including face recognition, mask wearing monitoring, crowd density monitoring, automatic body temperature measurement, etc. Some technological innovation companies have also launched new technologies and new products to help prevent and control the epidemic, such as rapid temperature measurement, mobile phone or voice control of elevators, contactless meal delivery, and disinfection robots.

Some Internet companies have also invested in the research and development of technology products related to epidemic monitoring, including the development of intelligent epidemic robots based on speech recognition technology and automatic human body temperature measurement and face recognition systems based on infrared thermal imaging and cloud face recognition technology. Some technological innovation companies have also launched new technologies and new products to help prevent and control the epidemic, such as rapid temperature measurement, mobile phone or voice control of elevators, contactless meal delivery, and disinfection robots.

In general, most of the existing technology products are only aimed at general public places, ignoring the important application site of the campus, making it difficult for the existing epidemic risk prevention and control technology products to play a key role in the campus epidemic prevention and control process.

SUMMARY OF THE INVENTION

In order to solve the above-mentioned problems, the present invention discloses a smart campus epidemic risk monitoring and early warning system and method, which aims to detect the diversion of crowds in the campus, analyze the travel characteristics of the campus crowd, design a variety of core monitoring functions, and provide accurate information for people traveling in the campus. The trajectory information, supplemented by the campus epidemic flow survey data, realizes the accurate prevention and control of the campus epidemic, and provides the campus administrators with accurate epidemic risk monitoring and early warning services to people, specific locations and times.

For achieving the above object, technical scheme of the present invention is as follows:

A smart campus epidemic risk monitoring and early warning system, comprising: an image data acquisition and processing module, an identity identification module, a target detection module, a trajectory identification module and an epidemic risk detection module;

The image data collection and processing module is used to collect the flow of the campus crowd, and pre-process according to the flow data, including data enhancement, data labeling and data set division. The image data collection and processing module includes the image data collection sub-module and the image data. preprocessing submodule;

The identity recognition module is used to recognize the target identity information based on pedestrian face recognition and pedestrian gesture recognition, and the identity recognition module includes a face recognition sub-module and a gesture recognition sub-module;

The target detection module is used to detect the pedestrian target state on the image data, and the target detection module includes a pedestrian detection sub-module, a pedestrian density detection sub-module and a mask detection sub-module;

The trajectory identification module is used to detect the action range and action mode of pedestrians, and the trajectory identification module includes a line crossing detection sub-module, a distance detection sub-module and a speed detection sub-module;

The epidemic risk detection module is used to obtain the data detected by the identity identification module, the target detection module and the trajectory identification module to identify the flow of pedestrians on campus, and to conduct research, judgment and early warning on the flow of the campus crowd. Monitoring sub-module, personnel flow statistics sub-module and communication risk warning sub-module.

Further, the image data collection sub-module collects the campus crowd moving video through external access to aerial photography equipment and several cameras, performs detection and tracking training on different samples in the collected image and video, and establishes an image data set;

The image data preprocessing sub-module is used to operate the collected images and videos, including flipping, rotating, cutting, splicing, and mosaic enhancement of images and videos, performing image morphological operations on images and videos, and using image pyramid technology to select modeling images. Video, use image annotation software tools to manually label images, determine the size of the a priori marker frame through clustering technology, obtain image data sets, and divide the collected data into training sets, validation sets and test sets according to the characteristics of the data set.

Further, the face recognition sub-module is externally connected to the campus face database, and a Cascade cascade classifier is constructed based on PCA dimensionality reduction and LDA feature extraction technology to identify the key point features of faces and their grayscale changes, and identify them according to different face features. , train the face recognition model, match the target in the campus face database through the recognized facial features, and carry out accurate identification;

The pose recognition sub-module is based on the image segmentation modeling of Masked R-CNN to segment pedestrian objects from images and videos, and uses the 2D key point detection algorithm to capture the key position information of the human body posture in the process of human behavior expression, so as to extract the key points of the human body. Characterization point to determine target identity information.

Further, the pedestrian detection sub-module is used to detect pedestrian targets in images and videos, and a multi-target detection model based on the Transformer architecture is constructed. Preliminary feature extraction, by constructing a Transformer network to accurately filter and nonlinearly transform the extracted pedestrian target features, and construct a prediction model by correlating feature information, classification labels and detection frame positions, and obtain the optimal target detection model through multiple rounds of iterative training. , so as to realize the detection mark of pedestrian target;

The pedestrian detection sub-module performs target tracking on the marked pedestrian target, constructs the TransTrack target tracking model, takes the pedestrian target feature map of the current frame as the input key, and uses the pedestrian target features of a series of past frames and a series of learnable queries as the input query. , by constructing a learnable query to detect new pedestrian targets and output detection bounding boxes, where the pedestrian target features of past frames are generated by past frame detection, which are used to locate existing objects in the current frame and output track boxes, through the detection frame Interact with the trajectory frame and compare and match to complete the output;

The pedestrian density detection sub-module uses a clustering algorithm to cluster the target marker positions of pedestrians, supplemented by a heat map to visually display the density of slow-moving traffic;

The mask detection sub-module is used to detect whether the campus pedestrians wear masks in the images and videos.

Further, the line crossing detection sub-module is used to automatically determine whether the pedestrian in the image and video has crossed the reference line, and if the line crossing occurs, an early warning message is automatically issued;

The distance detection sub-module is used to calculate the distance between two people in the image and video to determine whether there is a potential risk of epidemic transmission between them;

The speed detection sub-module is used to estimate the speed of the target pedestrian in the image and video.

Further, the high-risk area identification sub-module is used to identify high-risk areas of epidemic transmission in images and videos with large campus traffic and high trajectory overlap;

The illegal personnel monitoring sub-module is used to monitor the illegal behavior of campus personnel in images and videos in real time;

The personnel flow statistics sub-module is used to count the personnel flow in and out of different areas of the campus in images and videos;

The communication risk early warning sub-module is used to judge potential communication risks through images and videos, and provide campus administrators with rapid early warning services, including system early warning, SMS early warning and email early warning. The SMS early warning automatically triggers the early warning mechanism based on the researched and judged risk threshold, and sends early warning short messages to the campus administrator in time.

A smart campus epidemic risk monitoring and early warning method:

S1: Use the image data collection and processing module to collect the flow of people on campus, and perform preprocessing according to the flow data, including data enhancement, data labeling, and data set division and other preprocessing;

S2: use the identity recognition module to identify the target identity information based on pedestrian face recognition and pedestrian gesture recognition;

S3: Use the target detection module to detect the pedestrian target state on the image data;

S4: Use the trajectory identification module to detect the action range and action mode of pedestrians;

S5: Use the epidemic risk detection module to obtain the data detected by the identity identification module, the target detection module and the trajectory identification module to identify the pedestrian flow on campus, and conduct research, judgment and early warning on the flow of the campus crowd.

Further settings: Steps S1, S2, S3, S4, and S5 specifically include the following steps:

S1-1: Use the image data acquisition sub-module to collect the video of campus crowd flow through external access to aerial photography equipment and several cameras, perform detection and tracking training on different samples in the collected image and video, establish an image data set, and use the image data to preprocess The sub-module is used to operate the collected images and videos, including flipping, rotating, cutting, splicing and mosaicing the images and videos, performing image morphological operations on the images and videos, using the image pyramid technology to select the modeling images and videos, and using the image The labeling software tool manually labels the image, determines the size of the a priori labeling frame through clustering technology, obtains the image data set, and divides the collected data into training set, verification set and test set according to the characteristics of the data set;

S2-1: Use the face recognition sub-module to access the campus face database, build a Cascade cascade classifier based on PCA dimensionality reduction and LDA feature extraction technology, and identify the key point features of faces and their grayscale changes. Feature recognition, train a face recognition model, match the target in the campus face database through the recognized facial features, perform accurate identification, and use the gesture recognition sub-module based on the Masked R-CNN image segmentation modeling, to extract the image from the image. Segment pedestrian objects in the video, and use the 2D key point detection algorithm to capture the key position information of human body posture in the process of human behavior expression, so as to extract key human body characterization points and determine target identity information;

S3-1: Use the pedestrian detection sub-module to detect pedestrian targets in images and videos, build a multi-target detection model based on the Transformer architecture, and use the CNN convolutional neural network to detect pedestrian target features in video images by introducing a multi-head self-attention mechanism. Preliminary feature extraction, by constructing a Transformer network to accurately filter and nonlinearly transform the extracted pedestrian target features, and construct a prediction model by correlating feature information, classification labels and detection frame positions, and obtain the optimal target detection model through multiple rounds of iterative training. , so as to realize the detection and marking of pedestrian targets; the pedestrian detection sub-module performs target tracking on the marked pedestrian targets, constructs a TransTrack target tracking model, takes the pedestrian target feature map of the current frame as the input key, and uses the pedestrian target features of a series of past frames. And a series of learnable query as input query, by constructing learnable query, to detect new pedestrian target and output detection bounding box, among which, the pedestrian target feature of past frame is generated by past frame detection, used to locate the existing frame in the current frame. The object and the output trajectory frame are obtained through the intersection of the detection frame and the trajectory frame to complete the output. The pedestrian density detection sub-module uses the clustering algorithm to cluster the pedestrian target marker positions, supplemented by a heat map to visually display the slow traffic density. , the mask detection sub-module is used to detect whether campus pedestrians wear masks in the image and video;

S4-1: Use the line-crossing detection sub-module to automatically determine whether the pedestrian in the image and video crosses the reference line. If there is a line-crossing, an early warning message is automatically issued, and the distance detection sub-module is used to calculate the distance between two pedestrians in the image and video. Whether there is a potential risk of the spread of the epidemic, the speed detection sub-module is used to estimate the speed of the target pedestrian in the image and video;

S5-1: Use the high-risk area identification sub-module to identify the high-risk areas of the epidemic spread in the image and video with large flow of people and high trajectory overlap, and use the illegal personnel monitoring sub-module to monitor the illegal behavior of campus personnel in the image and video in real time. , using the personnel flow statistics sub-module to count the flow of people entering and exiting different areas of the campus in images and videos, and using the transmission risk early warning sub-module to judge potential transmission risks through images and videos, and provide campus administrators with rapid early warning services, including system early warning, SMS early warning and email early warning. Among them, the system early warning carries out risk early warning through the information pop-up box inside the system. The short message early warning automatically triggers the early warning mechanism according to the risk threshold value, and sends early warning short messages to the campus administrator in time. The email early warning is connected to the external mail system through the system. , write automatic mail sending program, carry out mail warning.

The beneficial effects of the present invention are:

The invention overcomes the disadvantages of the existing mainstream manual inspection methods and epidemic risk prevention and control technology products, such as time-consuming and labor-intensive, limited monitoring range and slow response speed. Epidemic flow survey data, realize the accurate prevention and control of the campus epidemic, and provide campus managers with accurate epidemic risk monitoring and early warning services to people, specific locations and times.

Description of drawings

1 is a schematic diagram of the overall module connection structure of the smart campus epidemic risk monitoring and early warning system provided by the present invention;

2 is a schematic diagram of the main steps of the smart campus epidemic risk monitoring and early warning method provided by the present invention;

3 is a schematic diagram of the specific steps of the smart campus epidemic risk monitoring and early warning method provided by the present invention;

FIG. 4 is a schematic diagram of a specific implementation process of the smart campus epidemic risk monitoring and early warning method provided by the present invention.

Detailed ways

The present invention will be further clarified below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and not to limit the scope of the present invention.

Referring to FIGS. 1 to 4 , an embodiment of the present invention provides a smart campus epidemic risk monitoring and early warning system. As shown in FIG. 1 , the system includes an image data acquisition and processing module, an identity recognition module, a target detection module, and a trajectory recognition module. and epidemic risk detection module.

The image data collection and processing module is used to collect the flow of the campus crowd, and preprocess it according to the flow data, including data enhancement, data labeling and data set division and other preprocessing;

It should be noted that the image data acquisition and processing module includes an image data acquisition sub-module and an image data preprocessing sub-module;

The image data acquisition sub-module collects the campus crowd moving video through external access to aerial photography equipment and several cameras, and performs detection and tracking training on different samples in the collected image and video to establish an image data set;

The image data preprocessing sub-module is used to operate the collected images and videos, including flipping, rotating, cutting, splicing, and mosaic enhancement of images and videos, performing image morphological operations on images and videos, and using image pyramid technology to select modeling images. Video, use image annotation software tools to manually label images, determine the size of the a priori marker frame through clustering technology, obtain image data sets, and divide the collected data into training sets, validation sets and test sets according to the characteristics of the data set.

The identity recognition module is used to recognize target identity information based on pedestrian face recognition and pedestrian gesture recognition.

It should be noted that the identity recognition module includes a face recognition sub-module and a gesture recognition sub-module.

The face recognition sub-module is connected to the campus face database, and the Cascade cascade classifier is constructed based on PCA dimensionality reduction and LDA feature extraction technology to identify the key point features of faces and their grayscale changes, and train people according to different face features. The face recognition model matches the target in the campus face database by identifying the facial features of the face to carry out accurate identification.

The pose recognition sub-module is based on the image segmentation modeling of Masked R-CNN to segment pedestrian objects from images and videos, and uses the 2D key point detection algorithm to capture the key position information of the human body posture in the process of human behavior expression, so as to extract the key points of the human body. Characterization point to determine target identity information.

The target detection module is used to detect the pedestrian target state on the image data.

It should be specified that the target detection module includes a pedestrian detection sub-module, a pedestrian density detection sub-module and a mask detection sub-module.

The pedestrian detection sub-module is used to detect pedestrian targets in images and videos, build a multi-target detection model based on the Transformer architecture, and introduce a multi-head self-attention mechanism. Feature extraction, by constructing Transformer network to accurately filter and non-linearly transform the extracted pedestrian target features, build a prediction model by correlating feature information, classification labels and detection frame positions, and obtain the optimal target detection model through multiple rounds of iterative training, So as to realize the detection mark of pedestrian target.

The pedestrian detection sub-module performs target tracking on the marked pedestrian target, constructs the TransTrack target tracking model, takes the pedestrian target feature map of the current frame as the input key, and uses the pedestrian target features of a series of past frames and a series of learnable queries as the input query. , by constructing a learnable query to detect new pedestrian targets and output detection bounding boxes, where the pedestrian target features of past frames are generated by past frame detection, which are used to locate existing objects in the current frame and output track boxes, through the detection frame Intersect with the track box and compare and match to complete the output.

The pedestrian density detection sub-module uses a clustering algorithm to cluster the target marker positions of pedestrians, supplemented by a heat map to visually display the density of slow-moving traffic.

The mask detection sub-module is used to detect whether the campus pedestrians wear masks in the images and videos.

The trajectory recognition module is used to detect the action range and action pattern of pedestrians.

It should be specified that the trajectory identification module includes a line crossing detection sub-module, a distance detection sub-module and a speed detection sub-module.

The line crossing detection sub-module is used to automatically judge whether the pedestrian in the image and video has crossed the baseline.

The distance detection sub-module is used to calculate the distance between two people in the image and video to determine whether there is a potential risk of epidemic transmission between the two.

The speed detection sub-module is used to estimate the speed of the target pedestrian in the image and video.

The epidemic risk detection module is used to obtain the data detected by the identity identification module, target detection module and trajectory identification module to identify the flow of pedestrians on campus, and to conduct research, judgment and early warning on the flow of people on campus. The epidemic risk detection module obtains the ID of the person from the identity identification module, obtains the category of the target (such as people, vehicles, etc.) and the location information in the image (the image coordinates of the target) from the target detection module, and obtains the image from the trajectory recognition. Time-transformed continuous image coordinate position information in .

It should be specified that the epidemic risk detection module includes a high-risk area identification sub-module, a violation personnel monitoring sub-module, a personnel flow statistics sub-module, and a transmission risk warning sub-module.

The high-risk area identification sub-module is used to identify high-risk areas for epidemic transmission in images and videos with large campus traffic and high trajectory overlap.

The illegal personnel monitoring sub-module is used to monitor the illegal behavior of campus personnel in images and videos in real time.

The personnel flow statistics sub-module is used to count the personnel flow in and out of different areas of the campus in the images and videos.

The communication risk warning sub-module is used to judge potential communication risks through images and videos, and provide campus administrators with rapid warning services, including system warnings, SMS warnings and email warnings. Among them, the system early warning carries out risk early warning through the information pop-up box inside the system, and the short message early warning automatically triggers the early warning mechanism according to the risk threshold value, and sends early warning short messages to the campus administrator in time. , to send an email alert. Specifically, when certain events are triggered through image judgment, such as not wearing a mask, pedestrians are too close, people crossing the line, or not belonging to campus personnel, the communication risk early warning sub-module will send early warning information to the system, mailbox and mobile phone.

As shown in Figure 2, a smart campus epidemic risk monitoring and early warning method:

S1: Use the image data collection and processing module to collect the flow of people on campus, and perform preprocessing according to the flow data, including data enhancement, data labeling, and data set division.

S2: Use the identity recognition module to identify the target identity information based on pedestrian face recognition and pedestrian gesture recognition.

S3: Use the target detection module to detect the pedestrian target state on the image data.

S4: Use the trajectory recognition module to detect the action range and action pattern of pedestrians.

S5: Use the epidemic risk detection module to obtain the data detected by the identity identification module, the target detection module and the trajectory identification module to identify the pedestrian flow on campus, and conduct research, judgment and early warning on the flow of the campus crowd.

As shown in Figure 3: further settings, steps S1, S2, S3, S4, S5 respectively include the following steps:

S1-1: Use the image data acquisition sub-module to collect the video of campus crowd flow through external access to aerial photography equipment and several cameras, perform detection and tracking training on different samples in the collected image and video, establish an image data set, and use the image data to preprocess The sub-module is used to operate the collected images and videos, including flipping, rotating, cutting, splicing and mosaicing the images and videos, performing image morphological operations on the images and videos, using the image pyramid technology to select the modeling images and videos, and using the image The annotation software tool manually annotates the image, determines the size of the a priori marker frame through clustering technology, obtains the image data set, and divides the collected data into training set, verification set and test set according to the characteristics of the data set.

S2-1: Use the face recognition sub-module to access the campus face database, build a Cascade cascade classifier based on PCA dimensionality reduction and LDA feature extraction technology, and identify the key point features of faces and their grayscale changes. Feature recognition, train a face recognition model, match the target in the campus face database through the recognized facial features, perform accurate identification, and use the gesture recognition sub-module based on the Masked R-CNN image segmentation modeling, to extract the image from the image. The pedestrian object is segmented in the video, and the 2D key point detection algorithm is used to capture the key position information of the human body posture in the process of human behavior expression, so as to extract the key human body characterization points and determine the target identity information.

S3-1: Use the pedestrian detection sub-module to detect pedestrian targets in images and videos, and build a multi-target detection model based on the Transformer architecture. By introducing a multi-head self-attention mechanism, the CNN convolutional neural network is used to detect pedestrian targets in video images. Preliminary feature extraction is performed on the features, and the extracted pedestrian target features are accurately filtered and nonlinearly transformed by constructing a Transformer network. By correlating feature information, classification labels and detection frame positions, a prediction model is constructed, and the optimal target is obtained through multiple rounds of iterative training. Detection model, so as to realize the detection and marking of pedestrian targets. The pedestrian detection sub-module performs target tracking on the marked pedestrian target, builds a TransTrack target tracking model, and uses the pedestrian target feature map of the current frame as the input key. A series of pedestrians in past frames The target feature and a series of learnable queries are input queries, and the learnable query is constructed to detect the new pedestrian target and output the detection bounding box. The existing object and output trajectory frame are output by the intersection of the detection frame and the trajectory frame, and the output is completed. The pedestrian density detection sub-module uses the clustering algorithm to cluster the pedestrian target marker positions, supplemented by a heat map to visualize the slow travel. Traffic density, the mask detection sub-module is used to detect whether campus pedestrians wear masks in images and videos.

S4-1: Use the line-crossing detection sub-module to automatically determine whether the pedestrian in the image and video crosses the reference line. If there is a line-crossing, an early warning message is automatically issued, and the distance detection sub-module is used to calculate the distance between two pedestrians in the image and video. Whether there is a potential risk of epidemic transmission among people, the speed detection sub-module is used to estimate the speed of the target pedestrian in the image and video.

S5-1: Use the high-risk area identification sub-module to identify the high-risk areas of the epidemic spread in the image and video with large flow of people and high trajectory overlap, and use the illegal personnel monitoring sub-module to monitor the illegal behavior of campus personnel in the image and video in real time. , using the personnel flow statistics sub-module to count the flow of people entering and exiting different areas of the campus in images and videos, and using the transmission risk early warning sub-module to judge potential transmission risks through images and videos, and provide campus administrators with rapid early warning services, including system early warning, SMS early warning and email early warning. Among them, the system early warning carries out risk early warning through the information pop-up box inside the system. The short message early warning automatically triggers the early warning mechanism according to the risk threshold value, and sends early warning short messages to the campus administrator in time. The email early warning is connected to the external mail system through the system. , write automatic mail sending program, carry out mail warning.

It should be noted that the above content only illustrates the technical idea of the present invention, and cannot limit the protection scope of the present invention. Several improvements and modifications can be made, which all fall within the protection scope of the claims of the present invention.

Claims (8)

1. A smart campus epidemic risk monitoring and early warning system, characterized in that, comprising: an image data acquisition and processing module, an identity identification module, a target detection module, a trajectory identification module and an epidemic risk detection module; The image data collection and processing module is used to collect the flow of the campus crowd, and pre-process according to the flow data, including data enhancement, data labeling, and data set division. The image data collection and processing module includes an image data collection sub-module and Image data preprocessing sub-module; The identity recognition module is used for identifying target identity information based on pedestrian face recognition and pedestrian gesture recognition, and the identity recognition module includes a face recognition submodule and a gesture recognition submodule; The target detection module is used to perform pedestrian target state detection on the image data, and the target detection module includes a pedestrian detection sub-module, a pedestrian density detection sub-module and a mask detection sub-module; The trajectory identification module is used to detect the action range and action mode of the pedestrian, and the trajectory identification module includes a line crossing detection sub-module, a distance detection sub-module and a speed detection sub-module; The epidemic risk detection module is used to obtain the data detected by the identity identification module, the target detection module and the trajectory identification module to identify the flow of pedestrians on campus, and to conduct research, judgment and early warning on the flow of the campus crowd. The epidemic risk detection module includes a high-risk area identification sub-module, Violating personnel monitoring sub-module, personnel flow statistics sub-module and dissemination risk warning sub-module. 2. smart campus epidemic risk monitoring and early warning system according to claim 1, is characterized in that: The image data collection sub-module collects the campus crowd moving video through external access to aerial photography equipment and several cameras, performs detection and tracking training on different samples inside the collected image and video, and establishes an image data set; The image data preprocessing sub-module is used to operate the collected images and videos, including flipping, rotating, cutting, splicing, and mosaic enhancement of the images and videos, performing image morphological operations on the images and videos, and selecting and constructing images by using the image pyramid technology. Model image videos, use image annotation software tools to manually annotate images, determine the size of a priori marked frame through clustering technology, obtain image data sets, and divide the collected data into training sets, validation sets, and test sets according to the characteristics of the data set. 3. smart campus epidemic risk monitoring and early warning system according to claim 1, is characterized in that: The face recognition sub-module is externally connected to the campus face database, and a Cascade cascade classifier is constructed based on PCA dimensionality reduction and LDA feature extraction technology to identify the key point features of faces and their grayscale changes. Train the face recognition model, and match the target in the campus face database through the recognized facial features to carry out accurate identification; The gesture recognition sub-module is based on the image segmentation modeling of Masked R-CNN to segment pedestrian objects from images and videos, and uses the 2D key point detection algorithm to capture the key position information of human body posture in the process of human behavior expression, so as to extract The key characterization points of the human body determine the target identity information. 4. smart campus epidemic risk monitoring and early warning system according to claim 1, is characterized in that: The pedestrian detection sub-module is used to detect pedestrian targets in images and videos, and build a multi-target detection model based on Transformer architecture. By introducing a multi-head self-attention mechanism, the CNN convolutional neural network is used to perform preliminary detection of pedestrian target features in video images. Feature extraction, by constructing Transformer network to accurately filter and non-linearly transform the extracted pedestrian target features, build a prediction model by correlating feature information, classification labels and detection frame positions, and obtain the optimal target detection model through multiple rounds of iterative training, So as to realize the detection mark of pedestrian target; The pedestrian detection sub-module performs target tracking on the marked pedestrian target, constructs a TransTrack target tracking model, takes the pedestrian target feature map of the current frame as the input key, and takes the pedestrian target features of a series of past frames and a series of learnable queries as the input key. Input query, by constructing a learnable query, to detect new pedestrian targets and output detection bounding boxes, where the pedestrian target features of past frames are generated by past frame detection, used to locate existing objects in the current frame and output track boxes, through The detection frame and the trajectory frame are crossed and matched to complete the output; The pedestrian density detection sub-module uses a clustering algorithm to cluster the target marker positions of pedestrians, and visually presents the density of slow-moving traffic by means of a heat map; The mask detection sub-module is used to detect whether campus pedestrians wear masks in the image and video. 5. smart campus epidemic risk monitoring and early warning system according to claim 1, is characterized in that: The line crossing detection sub-module is used to automatically judge whether the pedestrian in the image and video crosses the reference line, and if the line crossing occurs, an early warning message is automatically issued; The distance detection sub-module is used to calculate the distance between two people in the image and video, and to determine whether there is a potential epidemic transmission risk between the two people; The speed detection sub-module is used to estimate the speed of the target pedestrian in the image and video. 6. smart campus epidemic risk monitoring and early warning system according to claim 1, is characterized in that: The high-risk area identification sub-module is used to identify high-risk areas of epidemic spread in images and videos with large campus traffic and high trajectory overlap; The illegal personnel monitoring sub-module is used for real-time monitoring of the illegal behavior of campus personnel in images and videos; The personnel flow statistics sub-module is used to count the personnel flow in and out of different areas of the campus in the image and video; The communication risk early warning sub-module is used to judge potential communication risks through images and videos, and provide campus administrators with rapid early warning services, including system early warning, SMS early warning and email early warning. Early warning, SMS early warning automatically triggers the early warning mechanism based on the research and judgment risk threshold, and sends early warning short messages to campus administrators in time. 7. a smart campus epidemic risk monitoring and early warning method, is characterized in that, comprises the steps: S1: Use the image data collection and processing module to collect the flow of people on campus, and perform preprocessing according to the flow data, including data enhancement, data labeling, and data set division and other preprocessing; S2: use the identity recognition module to identify the target identity information based on pedestrian face recognition and pedestrian gesture recognition; S3: Use the target detection module to detect the pedestrian target state on the image data; S4: Use the trajectory identification module to detect the action range and action mode of pedestrians; S5: Use the epidemic risk detection module to obtain the data detected by the identity identification module, the target detection module and the trajectory identification module to identify the pedestrian flow on campus, and conduct research, judgment and early warning on the flow of the campus crowd. 8. The smart campus epidemic risk monitoring and early warning method according to claim 8, wherein the steps S1, S2, S3, S4, and S5 specifically include the following steps: S1-1: Use the image data acquisition sub-module to collect the video of campus crowd flow through external access to aerial photography equipment and several cameras, perform detection and tracking training on different samples in the collected image and video, establish an image data set, and use the image data to preprocess The sub-module is used to operate the collected images and videos, including flipping, rotating, cutting, splicing and mosaicing the images and videos, performing image morphological operations on the images and videos, using the image pyramid technology to select the modeling images and videos, and using the image The labeling software tool manually labels the image, determines the size of the a priori labeling frame through clustering technology, obtains the image data set, and divides the collected data into training set, verification set and test set according to the characteristics of the data set; S2-1: Use the face recognition sub-module to access the campus face database, build a Cascade cascade classifier based on PCA dimensionality reduction and LDA feature extraction technology, and identify the key point features of faces and their grayscale changes. Feature recognition, train a face recognition model, match the target in the campus face database through the recognized facial features, perform accurate identification, and use the gesture recognition sub-module based on the Masked R-CNN image segmentation modeling, to extract the image from the image. Segment pedestrian objects in the video, and use the 2D key point detection algorithm to capture the key position information of human body posture in the process of human behavior expression, so as to extract key human body characterization points and determine target identity information; S3-1: Use the pedestrian detection sub-module to detect pedestrian targets in images and videos, build a multi-target detection model based on the Transformer architecture, and use the CNN convolutional neural network to detect pedestrian target features in video images by introducing a multi-head self-attention mechanism. Preliminary feature extraction, by constructing a Transformer network to accurately filter and nonlinearly transform the extracted pedestrian target features, and construct a prediction model by correlating feature information, classification labels and detection frame positions, and obtain the optimal target detection model through multiple rounds of iterative training. , so as to realize the detection and marking of pedestrian targets, the pedestrian detection sub-module performs target tracking on the marked pedestrian targets, builds a TransTrack target tracking model, takes the pedestrian target feature map of the current frame as the input key, and uses the pedestrian target features of a series of past frames. And a series of learnable query as input query, by constructing learnable query, to detect new pedestrian target and output detection bounding box, among which, the pedestrian target feature of past frame is generated by past frame detection, used to locate the existing frame in the current frame. The object and the output trajectory frame are obtained through the intersection of the detection frame and the trajectory frame to complete the output. The pedestrian density detection sub-module uses the clustering algorithm to cluster the pedestrian target marker positions, supplemented by a heat map to visually display the slow traffic density. , the mask detection sub-module is used to detect whether campus pedestrians wear masks in the image and video; S4-1: Use the line-crossing detection sub-module to automatically determine whether the pedestrian in the image and video crosses the reference line. If there is a line-crossing, an early warning message is automatically issued, and the distance detection sub-module is used to calculate the distance between two pedestrians in the image and video. Whether there is a potential risk of the spread of the epidemic, the speed detection sub-module is used to estimate the speed of the target pedestrian in the image and video; S5-1: Use the high-risk area identification sub-module to identify the high-risk areas of the epidemic spread in the image and video with large flow of people and high trajectory overlap, and use the illegal personnel monitoring sub-module to monitor the illegal behavior of campus personnel in the image and video in real time. , using the personnel flow statistics sub-module to count the flow of people entering and exiting different areas of the campus in images and videos, and using the transmission risk early warning sub-module to judge potential transmission risks through images and videos, and provide campus administrators with rapid early warning services, including system early warning, SMS early warning and email early warning. Among them, the system early warning carries out risk early warning through the information pop-up box inside the system. The short message early warning automatically triggers the early warning mechanism according to the risk threshold value, and sends early warning short messages to the campus administrator in time. The email early warning is connected to the external mail system through the system. , write automatic mail sending program, carry out mail warning.

Images