High-speed railway perimeter intrusion risk monitoring method and system based on multi-target recognition

Abstract

The present invention discloses a high-speed railway perimeter intrusion risk monitoring method and system based on multi-target identification, which belongs to the field of railway security, wherein the method includes: deploying a distributed high-speed railway perimeter monitoring network, obtaining the real-time status of a sub-area through a camera unit, and obtaining a sub-area status result; when the status result is abnormal, recording the target sub-area, and obtaining a sub-area abnormal video stream; transmitting the sub-area status result and the sub-area abnormal video stream to a processing center, analyzing the sub-area status result and the sub-area abnormal video stream, and obtaining a terminal monitoring result; transmitting the terminal monitoring result to a railway control center for visual display. The present application solves the technical problems of incomplete monitoring of high-speed railway perimeter areas, low monitoring efficiency, and inaccurate risk judgment in the prior art, and achieves the technical effect of realizing real-time, comprehensive, accurate, and efficient monitoring of high-speed railway perimeter intrusion risks through distributed network monitoring and intelligent analysis and processing.

Description

Technical Field

[0001] The present invention relates to the field of railway security, and in particular to a high-speed railway perimeter intrusion risk monitoring method and system based on multi-target recognition. Background Art

[0002] With the rapid development of high-speed railway construction, the safe operation of high-speed railway as a public infrastructure is of great significance to people's livelihood. The important boundary area around the high-speed railway is an important protective barrier for stations and lines. Once various risk events such as illegal intrusion occur, it will seriously endanger the operational safety of the high-speed railway. Therefore, how to comprehensively, real-time and efficiently monitor and prevent various intrusion risks in the perimeter area of ​​the high-speed railway is an important part of ensuring the safety of the high-speed railway. At present, the monitoring and prevention methods of the perimeter area of ​​the high-speed railway are mainly to set up surveillance cameras and manual inspections. This monitoring method is unable to conduct all-round supervision of the perimeter environment. At the same time, it is easy to make omissions when relying on manual judgment, and it is difficult to achieve real-time, efficient and accurate intrusion detection. Summary of the invention

[0003] This application provides a high-speed railway perimeter intrusion risk monitoring method and system based on multi-target recognition, aiming to solve the technical problems of incomplete monitoring of high-speed railway perimeter areas, low monitoring efficiency and inaccurate risk judgment.

[0004] In view of the above problems, the present application provides a high-speed railway perimeter intrusion risk monitoring method and system based on multi-target recognition.

[0005] The first aspect disclosed in the present application provides a high-speed railway perimeter intrusion risk monitoring method based on multi-target identification, the method comprising: deploying the distributed high-speed railway perimeter monitoring network in the target high-speed railway perimeter area, setting the initial monitoring state of the target high-speed railway perimeter area for the distributed high-speed railway perimeter monitoring network; photographing the first target sub-area in the target high-speed railway perimeter area by a first camera unit in a first terminal monitoring node in the distributed high-speed railway perimeter monitoring network to obtain the real-time state of the first sub-area; based on the real-time state of the first sub-area, calling the first initial monitoring state of the first target sub-area in the initial monitoring state; comparing the real-time state of the first sub-area with the first initial monitoring state to obtain the state result of the first sub-area; when the first When there is an abnormality in the sub-area status result, the first recording unit in the first terminal monitoring node is activated to record the first target sub-area and obtain the abnormal video stream of the first sub-area; the first sub-area status result and the first sub-area abnormal video stream are transmitted to the first processing center in the first terminal monitoring node, and the first processing center analyzes the first sub-area status result and the first sub-area abnormal video stream to obtain the first terminal monitoring result, wherein the first terminal monitoring result includes a first intrusion subject set, a first intrusion route set, and a first intrusion risk level set; the first terminal monitoring result is transmitted to the railway control center in real time through the wireless ad hoc network, and the railway control center visualizes the first terminal monitoring result.

[0006] Another aspect disclosed in the present application provides a high-speed railway perimeter intrusion risk monitoring system based on multi-target recognition, the system comprising: a monitoring network deployment module, used to deploy the distributed high-speed railway perimeter monitoring network in the target high-speed railway perimeter area, and set the initial monitoring state of the target high-speed railway perimeter area for the distributed high-speed railway perimeter monitoring network; a real-time status acquisition module, used to photograph the first target sub-area in the target high-speed railway perimeter area through the first camera unit in the first terminal monitoring node in the distributed high-speed railway perimeter monitoring network, and obtain the real-time state of the first sub-area; an initial state retrieval module, used to retrieve the first initial monitoring state of the first target sub-area in the initial monitoring state based on the real-time state of the first sub-area; an area state result module, used to compare the real-time state of the first sub-area with the first initial monitoring state, and obtain the state result of the first sub-area an abnormal video stream module, used for activating the first recording unit in the first terminal monitoring node when the first sub-area status result is abnormal, recording the first target sub-area, and obtaining the first sub-area abnormal video stream; a terminal monitoring result module, used for transmitting the first sub-area status result and the first sub-area abnormal video stream to the first processing center in the first terminal monitoring node, analyzing the first sub-area status result and the first sub-area abnormal video stream based on the first processing center, and obtaining the first terminal monitoring result, wherein the first terminal monitoring result includes a first intrusion subject set, a first intrusion route set, and a first intrusion risk level set; a monitoring visualization module, used for transmitting the first terminal monitoring result to the railway control center in real time through the wireless ad hoc network, and the railway control center performs a visual display of the first terminal monitoring result.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] Due to the deployment of a distributed high-speed rail perimeter monitoring network in the target high-speed rail perimeter area, the network includes multiple terminal monitoring nodes to achieve all-round monitoring coverage of the perimeter area; the initial monitoring state of the area is set for the network as a reference state for judging regional abnormalities; the camera unit of the first monitoring node takes a picture of the sub-area it is responsible for, and obtains the real-time state information of the sub-area, which is the basis for comparative judgment; based on the real-time state of the sub-area, the corresponding initial state is retrieved for comparison to judge the deviation between the real-time state and the initial state; if there is an abnormality in the comparison result, the recording unit of the node is activated to record the sub-area, obtain the abnormal video stream, and provide the abnormal The processing center analyzes the status results and video streams to obtain monitoring results, including the intruder, route, risk level, etc., for intelligent identification and judgment; the monitoring results are transmitted to the control center in real time through the wireless network to achieve centralized management and visual display for convenient and rapid response, realizing a technical solution for comprehensive, real-time and intelligent monitoring of the intrusion risk in the perimeter area of ​​high-speed railways. This solves the technical problems of incomplete monitoring of the perimeter area of ​​high-speed railways, low monitoring efficiency and inaccurate risk judgment in the existing technology, and achieves the technical effect of realizing real-time, comprehensive, accurate and efficient monitoring of the intrusion risk in the perimeter of high-speed railways through distributed network monitoring and intelligent analysis and processing.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flow chart of a high-speed railway perimeter intrusion risk monitoring method based on multi-target recognition is provided for an embodiment of the present application;

[0011] Figure 2 A flow chart of deploying a target intrusion monitoring model in a high-speed railway perimeter intrusion risk monitoring method based on multi-target identification is provided for an embodiment of the present application;

[0012] Figure 3 A structural schematic diagram of a high-speed railway perimeter intrusion risk monitoring system based on multi-target recognition is provided for an embodiment of the present application.

[0013] Explanation of the reference numerals: monitoring network deployment module 11, real-time status acquisition module 12, initial status retrieval module 13, regional status result module 14, abnormal video stream module 15, terminal monitoring result module 16, monitoring visualization module 17. DETAILED DESCRIPTION

[0014] The overall idea of ​​the technical solution provided by this application is as follows:

[0015] The embodiment of the present application provides a method and system for monitoring the risk of high-speed railway perimeter intrusion based on multi-target identification. First, a distributed monitoring network is deployed in the target perimeter area, and the network includes multiple intelligent terminal nodes that can take photos and record videos to achieve all-round monitoring of the entire perimeter area. Then, an initial state is set for each sub-area, and the terminal node obtains the real-time state information of the corresponding sub-area in real time. By comparing and judging the real-time state with the initial state, abnormal situations can be quickly discovered and real-time monitoring can be achieved. Then, once an abnormality is determined, the recording unit of the terminal node is activated to record the abnormal area in detail. The state judgment results and the video are transmitted to the processing center in the terminal node for in-depth analysis, and key information such as the intruder and route are accurately obtained to achieve intelligent monitoring and analysis. Finally, the monitoring and analysis results are transmitted to the control center through the wireless network for centralized management and visual display so that security personnel can respond quickly, thereby achieving comprehensive, real-time, accurate and intelligent high-speed railway perimeter intrusion risk monitoring.

[0016] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification.

[0017] Embodiment 1

[0018] like Figure 1 As shown, an embodiment of the present application provides a high-speed railway perimeter intrusion risk monitoring method based on multi-target identification. The method is applied to a high-speed railway perimeter intrusion risk monitoring system based on multi-target identification. The system includes a distributed high-speed railway perimeter monitoring network, a wireless ad hoc network, and a railway control center. The distributed high-speed railway perimeter monitoring network includes multiple terminal monitoring nodes, and the terminal monitoring nodes include a camera unit, a video recording unit, and a processing center. The multiple terminal monitoring nodes are communicatively connected to the railway control center via a wireless ad hoc network.

[0019] Specifically, the embodiment of the present application discloses a high-speed railway perimeter intrusion risk monitoring method based on multi-target identification, which is applied to a high-speed railway perimeter intrusion risk monitoring system based on multi-target identification, the system includes a distributed high-speed railway perimeter monitoring network, a wireless ad hoc network, and a railway control center, and the distributed high-speed railway perimeter monitoring network is connected to the railway control center through the wireless ad hoc network. Among them, the distributed high-speed railway perimeter monitoring network is used to monitor the target high-speed railway perimeter area, and the distributed high-speed railway perimeter monitoring network includes a plurality of terminal monitoring nodes arranged in the target high-speed railway perimeter area; the wireless ad hoc network is used to realize the information transmission between the plurality of terminal monitoring nodes and the railway control center; the railway control center is used to receive the monitoring result information output by the plurality of terminal monitoring nodes and display and process it. Among them, the terminal monitoring node includes a camera unit, a video unit, and a processing center; the camera unit is used to collect images of the corresponding target sub-area of ​​the terminal monitoring node; the video unit is used to start the video recording of the target sub-area when the terminal monitoring node detects an abnormal situation; the processing center is used to analyze and process the monitoring information of the terminal monitoring node, and output the monitoring result.

[0020] High-speed railway perimeter intrusion risk monitoring methods include:

[0021] Deploy the distributed high-speed railway perimeter monitoring network in the target high-speed railway perimeter area, and set an initial monitoring state of the target high-speed railway perimeter area for the distributed high-speed railway perimeter monitoring network;

[0022] In the embodiment of the present application, the target high-speed rail perimeter area refers to the area around the high-speed rail line that needs to be monitored for intrusion, and the area is divided into multiple target sub-areas according to the direction of the high-speed rail line. In order to cover the target high-speed rail perimeter area to the greatest extent, multiple terminal monitoring nodes are arranged in the target area so that each terminal monitoring node can monitor a target sub-area. The camera unit of the terminal monitoring node is used to take static images, the video unit is used to capture dynamic videos, and the processing center analyzes and processes the acquired images and videos. Multiple terminal monitoring nodes constitute a distributed high-speed rail perimeter monitoring network to achieve monitoring of the entire target high-speed rail perimeter area. In addition, after the distributed high-speed rail perimeter monitoring network is laid out, the initial monitoring state of the target high-speed rail perimeter area is set for the monitoring network, including relevant information such as the topography, vegetation distribution, and obstacle arrangement of the target sub-area, which is used for subsequent intrusion risk assessment and detection analysis, laying the foundation for subsequent intrusion monitoring.

[0023] Photographing a first target sub-region in the target high-speed railway perimeter region by a first camera unit in a first terminal monitoring node in the distributed high-speed railway perimeter monitoring network to obtain a real-time status of the first sub-region;

[0024] In an embodiment of the present application, after the deployment of the distributed high-speed rail perimeter monitoring network is completed, each terminal monitoring node in the monitoring network is started for real-time monitoring. The first camera unit in the first terminal monitoring node is started to shoot the first target sub-area that it is responsible for monitoring, and the image of the first target sub-area is obtained in real time. Among them, the first terminal monitoring node is any terminal monitoring node among the multiple terminal monitoring nodes in the distributed high-speed rail perimeter monitoring network, which is responsible for monitoring the first target sub-area, and the first camera unit is an image acquisition device set on the first terminal monitoring node, which is a high-resolution camera or other equipment, and is responsible for acquiring images in the first target sub-area.

[0025] The first terminal monitoring node starts the first camera unit to automatically shoot the first target sub-area at a certain time interval to obtain real-time images of the sub-area, reflecting the topography, target object distribution and other information in the first target sub-area, forming the real-time status of the first sub-area, which is the basis for subsequent intrusion detection and risk assessment.

[0026] Based on the real-time state of the first sub-area, calling the first initial monitoring state of the first target sub-area in the initial monitoring state;

[0027] In the embodiment of the present application, after obtaining the first sub-region real-time state of the first target sub-region, based on the first sub-region real-time state, the first initial monitoring state corresponding to the first target sub-region is retrieved in the pre-set initial monitoring state. The first initial monitoring state is the initial state information corresponding to the first target sub-region.

[0028] When the obtained real-time status is subsequently analyzed, the real-time status needs to be compared with the initial status to determine whether an abnormal situation occurs in the target sub-area. Therefore, after obtaining the first sub-area real-time status of the first target sub-area, the corresponding first initial monitoring status is first retrieved to prepare for the subsequent comparative analysis. The information contained in the first initial monitoring status matches the information type and format collected when generating the first sub-area real-time status to ensure the comparability between the two states, thereby accurately determining abnormal situations.

[0029] By calling the first initial monitoring state corresponding to the real-time state of the first sub-area in the initial monitoring state, the corresponding association between the real-time state and the initial state is achieved, providing basic data for subsequent anomaly detection and intrusion risk assessment.

[0030] Compare the first sub-area real-time state with the first initial monitoring state to obtain a first sub-area state result;

[0031] In an embodiment of the present application, after obtaining the real-time state of the first sub-region of the first target sub-region and the corresponding first initial monitoring state. First, the images corresponding to the real-time state of the first sub-region and the first initial monitoring state are preprocessed, including image denoising, enhancement, etc., to improve the image quality. Then, the two images are aligned by image registration to ensure that the two images correspond to the same scene and perspective, and the image content is aligned. On the basis of image alignment, all pixels of the two images are traversed, and the color or gray value difference between the two at the same position is calculated, and the calculated difference value of each pixel is generated as a new difference image. Subsequently, according to the difference threshold value pre-set based on the monitoring sensitivity, threshold segmentation is performed on the difference image, and pixels with a difference value higher than the threshold are determined as abnormal pixels. Then, according to the connected area of ​​the abnormal pixel, the label of the abnormal area is given on the original image of the real-time state of the first sub-region, and the first sub-region state result is output according to the area of ​​the abnormal area and the pixel ratio, wherein the first sub-region state result is "normal" or "abnormal".

[0032] By obtaining the status result of the first sub-area, a basis is provided for judging whether there is intrusion in the target sub-area. The normal status result indicates that there is no intrusion risk and monitoring can continue; the abnormal status result will start the subsequent abnormal processing process, such as video recording, risk assessment, etc. By accurately and effectively detecting abnormal conditions in the target sub-area, a foundation is laid for subsequent intrusion monitoring and response.

[0033] When the first sub-area status result is abnormal, activating the first recording unit in the first terminal monitoring node to record the first target sub-area and obtain an abnormal video stream of the first sub-area;

[0034] In the embodiment of the present application, when the abnormal state result is obtained by comparing and analyzing the real-time state of the first sub-region and the first initial monitoring state, it indicates that there may be an intrusion risk in the first target sub-region corresponding to the real-time state of the first sub-region. In order to further analyze the risk status, it is necessary to obtain the dynamic abnormal information of the sub-region.

[0035] When an abnormality is detected in the status result of the first sub-area, the first recording unit in the corresponding first terminal monitoring node is immediately activated to start dynamic recording of the first target sub-area to obtain video information of the abnormal situation, i.e., the abnormal video stream of the first sub-area. The first recording unit uses a high-definition digital camera and other equipment to perform dynamic recording without blind spots; the obtained abnormal video stream of the first sub-area contains a real-time image sequence, which can record the entire dynamic process of the possible abnormal situation of the first target sub-area.

[0036] By obtaining the abnormal video stream of the first sub-area, information is provided for locating the abnormal source, determining the intrusion method, and evaluating the severity of the incident, thereby ensuring comprehensive monitoring of abnormal situations and providing a basis for intrusion risk monitoring.

[0037] The first sub-region status result and the first sub-region abnormal video stream are transmitted to a first processing center in the first terminal monitoring node, and the first processing center analyzes the first sub-region status result and the first sub-region abnormal video stream to obtain a first terminal monitoring result, wherein the first terminal monitoring result includes a first intrusion subject set, a first intrusion route set, and a first intrusion risk level set;

[0038] In an embodiment of the present application, after obtaining the first sub-region status result and the first sub-region abnormal video stream, the two are input into the first processing center of the corresponding first terminal monitoring node for comprehensive analysis to obtain detailed intrusion monitoring results. The first processing center has an embedded target intrusion monitoring model, which can perform intelligent analysis and processing on the sub-region status results and the sub-region abnormal video stream. The first processing center performs target detection, tracking, and behavior analysis on the first sub-region status results and the first sub-region abnormal video stream to locate the intrusion subject, determine the intrusion route, and assess the danger level, and generate the monitoring results of the first terminal monitoring node. The results include the first intrusion subject set, the first intrusion route set, and the first intrusion risk level set, which respectively reflect the intrusion objects, intrusion routes, and risk judgment results of abnormal video monitoring, and provide a basis for subsequent decision-making.

[0039] The monitoring result of the first terminal is transmitted to the railway control center in real time via the wireless ad hoc network, and the railway control center displays the monitoring result of the first terminal in a visual manner.

[0040] In an embodiment of the present application, after the first processing center of the first terminal monitoring node obtains the first terminal monitoring result, the first terminal monitoring result is transmitted to the railway control center in real time through the wireless ad hoc network. Among them, the wireless ad hoc network can enable the terminal monitoring nodes in the distributed high-speed rail perimeter monitoring network to form an autonomous network, establish a flexible and reliable wireless relay link, and realize multi-hop wireless transmission of monitoring data. The first terminal monitoring result is transmitted to the monitoring station of the railway control center through the wireless ad hoc network. The monitoring station integrates an information visualization platform, which can summarize and integrate the received terminal monitoring results for vivid visualization. The visualization results can intuitively display the abnormal detection and intrusion risk distribution in the surrounding area of ​​the target high-speed rail line, and reflect the monitoring information in the form of graphics, animation, etc., which is convenient for security personnel to quickly grasp the abnormal situation and risk dynamics, make corresponding decisions, and realize the information interconnection and sharing between the monitoring network and the control center, so that the monitoring results can be fed back to the decision-maker in real time, which is conducive to improving the security response efficiency of railway facilities.

[0041] Furthermore, the embodiment of the present application also includes:

[0042] Delimiting a perimeter intrusion prevention area for the target high-speed railway perimeter area, determining an intrusion risk level based on the perimeter intrusion prevention area, and obtaining a regional intrusion risk distribution map;

[0043] Dividing the regional intrusion risk distribution map based on the monitoring areas of the multiple terminal monitoring nodes, and obtaining sub-region intrusion risk distribution maps corresponding to the multiple terminal monitoring nodes;

[0044] Acquire a first sub-area intrusion risk distribution map based on the first terminal monitoring node;

[0045] The intrusion risk level of the intruder is determined according to the intrusion risk distribution map of the first sub-region.

[0046] In a preferred embodiment, first, within the target high-speed rail perimeter area, the railway perimeter intrusion prevention area is delineated based on factors such as the distribution of railway facilities, topography, and traffic conditions to obtain a perimeter intrusion prevention area. This area includes strip areas on both sides of the railway line, as well as other sensitive areas connected to or adjacent to the area. Secondly, within the delineated perimeter intrusion prevention area, different intrusion risk levels are divided according to conditions such as terrain features, monitoring deployment, and response time at different locations. The closer to the railway facilities, the higher the corresponding intrusion risk level. Then, based on the distribution of intrusion risk levels, a regional intrusion risk distribution map is formed, which intuitively displays the intrusion risk levels at different locations within the perimeter intrusion prevention area, providing a basis for refining the intrusion risk levels according to geographical locations, and realizing refined risk analysis.

[0047] After obtaining the overall intrusion risk distribution map of the target high-speed railway perimeter area, the intrusion risk distribution map is finely distributed to each terminal monitoring node to guide the risk analysis of each node. First, according to the location and monitoring range of the multiple terminal monitoring nodes that have been deployed, the monitoring sub-areas of each terminal monitoring node are reasonably planned in the target high-speed railway perimeter area; on the regional intrusion risk distribution map, the monitoring sub-areas of the terminal monitoring nodes are divided so that each sub-area corresponds to a terminal monitoring node, and the intrusion risk level information in the sub-area is extracted. In this way, each terminal monitoring node obtains a corresponding sub-area intrusion risk distribution map, which contains the intrusion risk level information of each location in the sub-area that the node is responsible for monitoring. According to the sub-area intrusion risk distribution map, each terminal monitoring node can clearly understand the intrusion risk situation of different locations in its own monitoring area, providing a basis for the intrusion risk judgment of each node. Through the subdivision of the corresponding regional information, the refinement of the risk analysis of the monitoring network is realized, which helps to improve the pertinence and effectiveness of protection.

[0048] After the regional invasion risk distribution map is divided, each terminal monitoring node obtains the corresponding sub-region invasion risk distribution map. Taking the first terminal monitoring node as an example, according to the specific position of the node in the target high-speed rail perimeter area, the sub-region corresponding to the first terminal monitoring node in the regional invasion risk distribution map is extracted to obtain the first sub-region invasion risk distribution map. The sub-region invasion risk distribution map contains the invasion risk level of each geographical location point in the sub-region that the first terminal monitoring node is responsible for monitoring, and the different risk level areas on the map have clear boundaries. After the first terminal monitoring node obtains the invasion risk distribution map of the corresponding sub-region, when the first terminal monitoring node detects an intrusion subject through the real-time status of the first sub-region, the position coordinates of the subject can be determined. On the first sub-region invasion risk distribution map, according to the position coordinates of the intrusion subject, it is determined which risk level sub-region in the risk level distribution map the area where the intrusion subject is located belongs. In this way, the invasion risk level of the intrusion subject is the level of the risk sub-region, thereby accurately obtaining the invasion risk level of the intrusion subject. When the first terminal monitoring node monitors multiple intrusion subjects, a first intrusion subject set is formed, and each intrusion subject in the first intrusion subject set has a corresponding invasion risk level, and a first intrusion risk level set is obtained. The first intrusion risk level set is dynamically changed according to the real-time position coordinates of each intrusion subject in the first intrusion subject set.

[0049] Further, such as Figure 2 As shown, the embodiment of the present application also includes:

[0050] Collecting historical image data of the high-speed railway perimeter, marking the intrusion subjects in the historical images of the high-speed railway perimeter, and obtaining a perimeter intrusion image data set;

[0051] Based on the YOLOv3 code framework, load the pre-trained weights, use the perimeter intrusion image dataset to train the model, and obtain the initial intrusion monitoring model;

[0052] Exporting an initial intrusion monitoring model, converting it into a node deployment format, optimizing the initial intrusion monitoring model based on the node deployment format, and obtaining a target intrusion monitoring model;

[0053] The target intrusion monitoring model is deployed in the processing center of multiple terminal monitoring nodes, and is connected to the camera unit and the video recording unit through a wired communication.

[0054] In a preferred embodiment, first, by collecting historical surveillance images and video materials of the high-speed rail perimeter area, the high-speed rail perimeter historical images that clearly contain the intrusion target are selected to obtain the high-speed rail perimeter historical image data, which covers various scenes in the perimeter area. Secondly, using image annotation software, professional annotators accurately mark the location of the intrusion target in these high-speed rail perimeter historical images, such as intruders, vehicles, explosives, etc., to obtain a large number of high-speed rail perimeter images with intrusion target annotations to form a perimeter intrusion image data set. Then, YOLOv3 is used as the model framework. On the basis of the YOLOv3 framework, the pre-trained weights are loaded, that is, the model weight parameters trained in advance on a large-scale image data set are used to initialize the model, so that the model can converge quickly. Subsequently, the generated perimeter intrusion image data set is used as a new data set, and the model training is performed based on the strategy of transfer learning to obtain an initial intrusion monitoring model suitable for high-speed rail perimeter intrusion monitoring.

[0055] Next, in order to make the trained initial intrusion monitoring model run efficiently in the terminal monitoring node, the initial intrusion monitoring model is first exported to a general format, such as ONNX format, and then converted into a model format suitable for deployment in the terminal monitoring node, such as TensorRT model, according to the hardware requirements of the terminal monitoring node. After obtaining the initial intrusion monitoring model in the node deployment format, the model structure is streamlined and the model size is reduced by using optimization technologies such as model compression and quantization, and the model is optimized to generate a target intrusion monitoring model with higher operating efficiency. The target model not only retains the monitoring effect of the initial intrusion monitoring model, but also has the optimization characteristics of being deployed on the terminal monitoring node to realize real-time automated intrusion monitoring in the high-speed rail perimeter area. After obtaining the target intrusion detection model optimized for node deployment, it is deployed to multiple actual terminal monitoring nodes, and the target intrusion detection model is loaded in the processing center of multiple terminal monitoring nodes in the distributed high-speed rail perimeter monitoring network, and connected to the processing center to realize the efficient operation of the model. In addition, ensure that there is a wired communication connection between the processing center and the camera unit and video recording unit of the terminal monitoring node, so that the sub-area status results and the sub-area abnormal video stream are transmitted to the processing center in real time for intrusion judgment.

[0056] Furthermore, the embodiment of the present application also includes:

[0057] After transmitting the first sub-region status result and the first sub-region abnormal video stream to the first processing center in the first terminal monitoring node, extracting the first sub-region real-time image corresponding to the first sub-region real-time status from the first sub-region status result;

[0058] Inputting the first sub-region real-time image into the target intrusion monitoring model to obtain a first intrusion subject set, wherein different intrusion subjects in the first intrusion subject set have a subject unique identifier and a starting coordinate point, and the starting coordinate point has a coordinate point time identifier;

[0059] Add the first intrusion subject set to the first terminal monitoring result.

[0060] In a preferred embodiment, when the first processing center receives the first sub-region status result and the first sub-region abnormal video stream, first, a static screen image corresponding to the real-time state of the first sub-region is extracted from the first sub-region status result, that is, the first sub-region real-time image, to detect the intrusion target and provide static picture data support for subsequent intrusion identification and monitoring. Then, the first sub-region real-time image is input into the target intrusion detection model pre-deployed in the terminal monitoring node, and the model identifies each intrusion instance in the image, such as an illegal intruder, a vehicle, etc.; for each identified intrusion instance, the model will generate a subject unique identifier, and at the same time give the coordinate position of the instance as the starting coordinate point, and the image acquisition time corresponding to the coordinate position as the coordinate point time identifier. All detected intrusion subjects and their subject unique identifiers, starting coordinate points and coordinate point time identifiers are integrated to form a first intrusion subject set, complete the monitoring and labeling of the intrusion subject, obtain the key initial information of the intrusion subject, and lay the foundation for subsequent behavior analysis. After that, the first intrusion subject set is added to the first terminal monitoring result to lay the foundation for subsequent tracking identification and risk analysis.

[0061] Furthermore, the embodiment of the present application also includes:

[0062] After obtaining the first intrusion subject set, evenly-intervally extracting image frames from the first sub-region abnormal video stream to obtain a first abnormal image frame set, wherein each abnormal image frame in the first abnormal image frame set carries an image frame time identifier;

[0063] Based on the image frame time identifier, the abnormal image frames in the first abnormal image frame set are sequentially input into the target intrusion monitoring model in chronological order, and the first tracking monitoring results are sequentially output;

[0064] The first tracking and monitoring result includes movement coordinate points of different intrusion entities, and the movement coordinate points are provided with coordinate point time identifiers, and the coordinate point time identifiers are consistent with the image frame time identifiers.

[0065] In a preferred embodiment, after obtaining the first intrusion subject set, uniform image extraction is performed on the first sub-region abnormal video stream obtained for the first target sub-region, that is, a single frame image is extracted from the first sub-region abnormal video stream at a fixed time interval, and the selected time interval is determined according to the video frame rate and actual needs. After extraction, a first abnormal image frame set containing multiple frames of images is formed, which is used to input the target intrusion monitoring model to track the intrusion target. At the same time, when extracting each frame of abnormal image, the time mark corresponding to the frame is brought, so that each frame of the first abnormal image frame set has a corresponding image frame time mark.

[0066] Then, after obtaining the first abnormal image frame set containing the image frame time mark, the abnormal image frames in the image frame set are input into the target intrusion monitoring model one by one according to the time mark order, and the model will identify the new coordinate positions of different intrusion subjects in each frame. By sequentially processing the frame images, the model analyzes the movement trajectory of the intrusion subject at different times, realizes its tracking and monitoring, and outputs the first tracking and monitoring result, which contains the coordinate position of each intrusion target in different frames, and carries the coordinate point time mark of the coordinate, which is consistent with the corresponding image frame time mark. In this way, the coordinate movement information of each intrusion subject in the abnormal video stream of the first sub-area is obtained, which provides information for completing the route tracking and identification of the intrusion subject.

[0067] Furthermore, the embodiment of the present application also includes:

[0068] After obtaining the first tracking and monitoring result, matching the first intrusion subject set with the same intrusion subject in the first tracking and monitoring result based on the subject unique identifier to obtain an intrusion coordinate point set corresponding to the same intrusion subject;

[0069] Based on the intrusion coordinate point set, taking the same intrusion subject as the fitting subject, taking the starting coordinate point as the intrusion starting point, taking the moving coordinate point as the intrusion moving point, performing curve fitting according to the coordinate point time mark, and acquiring a first intrusion route set;

[0070] Add the first intrusion route set to the first terminal monitoring result.

[0071] In a feasible implementation, based on the image frame time identifier, the abnormal image frames in the first abnormal image frame set are sequentially input into the target intrusion monitoring model in chronological order, and each abnormal image frame can obtain the corresponding first tracking monitoring result. After obtaining the first tracking monitoring result, the first intrusion subject set is matched with the same intrusion subject in the first tracking monitoring result according to the subject unique identifier, and the intrusion coordinate point set corresponding to each intrusion subject is obtained, including its starting coordinate point and the corresponding tracking coordinate points in each first tracking monitoring result, providing a data basis for the generation of the intrusion route.

[0072] Then, each time a first tracking monitoring result is obtained, the intrusion route of each intrusion subject is obtained based on the previous intrusion coordinate point set to form a dynamic first intrusion route set. Specifically, the intrusion subject is set as the fitting subject, the starting coordinate point of the intrusion subject is used as the intrusion starting point, and the tracking coordinate point in the first tracking monitoring result is used as the intrusion moving point. According to the coordinate point time mark corresponding to each coordinate, a curve fitting method is used to generate the intrusion route from the starting point to each moving point. Finally, each intrusion subject fits an intrusion route representing its motion trajectory, and the intrusion route set of all intrusion subjects constitutes the first intrusion route set. Among them, the generation of the intrusion route is dynamic and synchronous, that is, while the abnormal video stream of the first sub-region is being recorded, the abnormal video stream of the first sub-region is synchronously transmitted to the first processing center for image frame extraction, and the image frame is input into the target intrusion monitoring model for identification to obtain the first tracking monitoring result, and then the curve fitting is performed according to the moving coordinate point in the first tracking monitoring result to achieve real-time response to the intrusion.

[0073] Furthermore, the embodiment of the present application also includes:

[0074] The first terminal monitoring node also includes a first infrared detection unit;

[0075] When there is an abnormality in the status result of the first sub-area and the real-time light intensity of the first target sub-area is lower than the preset light intensity, the first infrared recording unit in the first terminal monitoring node is activated to record the first target sub-area and obtain the abnormal video stream of the first sub-area.

[0076] In a preferred embodiment, in order to enhance the monitoring capability of the system at night or under poor lighting conditions, the first terminal monitoring node includes a first infrared detection unit in the infrared band in addition to the video recording unit in the visible light band. The infrared detection unit is composed of infrared cameras and other devices, which can sense and collect image and video information in the infrared band.

[0077] When the first sub-area status result detects that there is an abnormality in the first target sub-area, if the real-time light intensity of the target sub-area is lower than the preset light intensity, it indicates that the ambient light in the target sub-area is dim, belonging to a night or dark scene, and it is not suitable to continue to use the visible light recording unit to collect information. In this case, the first infrared detection unit in the first terminal monitoring node is started to record the first target sub-area under the infrared spectrum. The first infrared detection unit can collect thermal imaging information and video data in the target sub-area under weak light or no light conditions, obtain the performance of abnormal conditions in the infrared band, and generate a first infrared abnormal video stream as the first sub-area abnormal video stream, so as to improve the system's ability to handle complex lighting conditions and enhance the all-weather monitoring effect.

[0078] In summary, the high-speed railway perimeter intrusion risk monitoring method based on multi-target recognition provided by the embodiment of the present application has the following technical effects:

[0079] A distributed high-speed rail perimeter monitoring network is deployed in the target high-speed rail perimeter area, and the initial monitoring state of the target high-speed rail perimeter area is set for the distributed high-speed rail perimeter monitoring network to achieve comprehensive monitoring coverage of the entire perimeter area. The first target sub-area in the target high-speed rail perimeter area is photographed by the first camera unit in the first terminal monitoring node in the distributed high-speed rail perimeter monitoring network to obtain the real-time state of the first sub-area, providing a benchmark for subsequent abnormal judgment. Based on the real-time state of the first sub-area, the first initial monitoring state of the first target sub-area is retrieved in the initial monitoring state; the real-time state of the first sub-area is compared with the first initial monitoring state to obtain the state result of the first sub-area and determine whether the current situation is abnormal. When there is an abnormality in the state result of the first sub-area, the first recording unit in the first terminal monitoring node is activated to record the first target sub-area, obtain the abnormal video stream of the first sub-area, and provide data support for analyzing the abnormality. The first sub-area status result and the first sub-area abnormal video stream are transmitted to the first processing center in the first terminal monitoring node, and the first processing center analyzes the first sub-area status result and the first sub-area abnormal video stream to obtain the first terminal monitoring result, thereby realizing intelligent monitoring; the first terminal monitoring result is transmitted to the railway control center in real time through the wireless ad hoc network, and the railway control center visualizes the first terminal monitoring result, thereby realizing comprehensive, real-time and accurate intelligent monitoring of the intrusion risk of the high-speed railway perimeter area.

[0080] Embodiment 2

[0081] Based on the same inventive concept as the high-speed railway perimeter intrusion risk monitoring method based on multi-target recognition in the aforementioned embodiment, Figure 3As shown, the embodiment of the present application provides a high-speed railway perimeter intrusion risk monitoring system based on multi-target identification, the system includes a distributed high-speed railway perimeter monitoring network, a wireless ad hoc network, and a railway control center, the distributed high-speed railway perimeter monitoring network includes a plurality of terminal monitoring nodes, the terminal monitoring nodes include a camera unit, a video unit, and a processing center, and the plurality of terminal monitoring nodes are connected to the railway control center through a wireless ad hoc network. The method includes:

[0082] A monitoring network deployment module 11 is used to deploy the distributed high-speed railway perimeter monitoring network in the target high-speed railway perimeter area and set an initial monitoring state of the target high-speed railway perimeter area for the distributed high-speed railway perimeter monitoring network;

[0083] A real-time status acquisition module 12 is used to photograph a first target sub-area in the target high-speed railway perimeter area through a first camera unit in a first terminal monitoring node in the distributed high-speed railway perimeter monitoring network to acquire a real-time status of the first sub-area;

[0084] An initial state retrieving module 13, configured to retrieve a first initial monitoring state of the first target sub-area in the initial monitoring state based on the real-time state of the first sub-area;

[0085] A region status result module 14, configured to compare the first sub-region real-time status with the first initial monitoring status to obtain a first sub-region status result;

[0086] The abnormal video stream module 15 is used to activate the first recording unit in the first terminal monitoring node to record the first target sub-area and obtain the abnormal video stream of the first sub-area when the first sub-area status result is abnormal;

[0087] The terminal monitoring result module 16 is used to transmit the first sub-region status result and the first sub-region abnormal video stream to the first processing center in the first terminal monitoring node, and analyze the first sub-region status result and the first sub-region abnormal video stream based on the first processing center to obtain a first terminal monitoring result, wherein the first terminal monitoring result includes a first intrusion subject set, a first intrusion route set, and a first intrusion risk level set;

[0088] The monitoring visualization module 17 is used to transmit the monitoring result of the first terminal to the railway control center in real time through the wireless ad hoc network, and the railway control center visualizes the monitoring result of the first terminal.

[0089] Furthermore, the embodiment of the present application also includes a risk level classification module, and the risk level classification module includes the following execution steps:

[0090] Delimiting a perimeter intrusion prevention area for the target high-speed railway perimeter area, determining an intrusion risk level based on the perimeter intrusion prevention area, and obtaining a regional intrusion risk distribution map;

[0091] Dividing the regional intrusion risk distribution map based on the monitoring areas of the multiple terminal monitoring nodes, and obtaining sub-region intrusion risk distribution maps corresponding to the multiple terminal monitoring nodes;

[0092] Acquire a first sub-area intrusion risk distribution map based on the first terminal monitoring node;

[0093] The intrusion risk level of the intruder is determined according to the intrusion risk distribution map of the first sub-region.

[0094] Furthermore, the terminal monitoring result module 16 includes the following execution steps:

[0095] Collecting historical image data of the high-speed railway perimeter, marking the intrusion subjects in the historical images of the high-speed railway perimeter, and obtaining a perimeter intrusion image data set;

[0096] Based on the YOLOv3 code framework, load the pre-trained weights, use the perimeter intrusion image dataset to train the model, and obtain the initial intrusion monitoring model;

[0097] Exporting an initial intrusion monitoring model, converting it into a node deployment format, optimizing the initial intrusion monitoring model based on the node deployment format, and obtaining a target intrusion monitoring model;

[0098] The target intrusion monitoring model is deployed in the processing center of multiple terminal monitoring nodes, and is connected to the camera unit and the video recording unit through a wired communication.

[0099] Furthermore, the terminal monitoring result module 16 further includes the following execution steps:

[0100] After transmitting the first sub-region status result and the first sub-region abnormal video stream to the first processing center in the first terminal monitoring node, extracting the first sub-region real-time image corresponding to the first sub-region real-time status from the first sub-region status result;

[0101] Inputting the first sub-region real-time image into the target intrusion monitoring model to obtain a first intrusion subject set, wherein different intrusion subjects in the first intrusion subject set have a subject unique identifier and a starting coordinate point, and the starting coordinate point has a coordinate point time identifier;

[0102] Add the first intrusion subject set to the first terminal monitoring result.

[0103] Furthermore, the terminal monitoring result module 16 further includes the following execution steps:

[0104] After obtaining the first intrusion subject set, evenly-intervally extracting image frames from the first sub-region abnormal video stream to obtain a first abnormal image frame set, wherein each abnormal image frame in the first abnormal image frame set carries an image frame time identifier;

[0105] Based on the image frame time identifier, the abnormal image frames in the first abnormal image frame set are sequentially input into the target intrusion monitoring model in chronological order, and the first tracking monitoring results are sequentially output;

[0106] The first tracking and monitoring result includes movement coordinate points of different intrusion entities, and the movement coordinate points are provided with coordinate point time identifiers, and the coordinate point time identifiers are consistent with the image frame time identifiers.

[0107] Furthermore, the terminal monitoring result module 16 further includes the following execution steps:

[0108] After obtaining the first tracking and monitoring result, matching the first intrusion subject set with the same intrusion subject in the first tracking and monitoring result based on the subject unique identifier to obtain an intrusion coordinate point set corresponding to the same intrusion subject;

[0109] Based on the intrusion coordinate point set, taking the same intrusion subject as the fitting subject, taking the starting coordinate point as the intrusion starting point, taking the moving coordinate point as the intrusion moving point, performing curve fitting according to the coordinate point time mark, and acquiring a first intrusion route set;

[0110] Add the first intrusion route set to the first terminal monitoring result.

[0111] Furthermore, the embodiment of the present application also includes an infrared detection module, which includes the following execution steps:

[0112] The first terminal monitoring node also includes a first infrared detection unit;

[0113] When there is an abnormality in the status result of the first sub-area and the real-time light intensity of the first target sub-area is lower than the preset light intensity, the first infrared recording unit in the first terminal monitoring node is activated to record the first target sub-area and obtain the abnormal video stream of the first sub-area.

[0114] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0115] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to multiple elements that can be selected individually or in full. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A high-speed railway perimeter intrusion risk monitoring method based on multi-target recognition, characterized in that: The method is applied to a high-speed railway perimeter intrusion risk monitoring system based on multi-target recognition, the system comprising a distributed high-speed railway perimeter monitoring network, a wireless ad hoc network, and a railway control center, the distributed high-speed railway perimeter monitoring network comprising a plurality of terminal monitoring nodes, the terminal monitoring nodes comprising a camera unit, a video recording unit, and a processing center, the plurality of terminal monitoring nodes being connected to the railway control center through the wireless ad hoc network, the method comprising: Deploy the distributed high-speed railway perimeter monitoring network in the target high-speed railway perimeter area, and set an initial monitoring state of the target high-speed railway perimeter area for the distributed high-speed railway perimeter monitoring network; Photographing a first target sub-region in the target high-speed railway perimeter region by a first camera unit in a first terminal monitoring node in the distributed high-speed railway perimeter monitoring network to obtain a real-time status of the first sub-region; Based on the real-time state of the first sub-area, calling the first initial monitoring state of the first target sub-area in the initial monitoring state; Compare the first sub-area real-time state with the first initial monitoring state to obtain a first sub-area state result; When the first sub-area status result is abnormal, activating the first recording unit in the first terminal monitoring node to record the first target sub-area and obtain an abnormal video stream of the first sub-area; The first sub-region status result and the first sub-region abnormal video stream are transmitted to a first processing center in the first terminal monitoring node, and the first processing center analyzes the first sub-region status result and the first sub-region abnormal video stream to obtain a first terminal monitoring result, wherein the first terminal monitoring result includes a first intrusion subject set, a first intrusion route set, and a first intrusion risk level set; The monitoring result of the first terminal is transmitted to the railway control center in real time via the wireless ad hoc network, and the railway control center displays the monitoring result of the first terminal in a visual manner.

2. The method according to claim 1, characterized in that The method further comprises: Delimiting a perimeter intrusion prevention area for the target high-speed railway perimeter area, determining an intrusion risk level based on the perimeter intrusion prevention area, and obtaining a regional intrusion risk distribution map; Dividing the regional intrusion risk distribution map based on the monitoring areas of the multiple terminal monitoring nodes, and obtaining sub-region intrusion risk distribution maps corresponding to the multiple terminal monitoring nodes; Acquire a first sub-area intrusion risk distribution map based on the first terminal monitoring node; The intrusion risk level of the intruder is determined according to the intrusion risk distribution map of the first sub-region.

3. The method according to claim 1, characterized in that The method comprises: Collecting historical image data of the high-speed railway perimeter, marking the intrusion subjects in the historical images of the high-speed railway perimeter, and obtaining a perimeter intrusion image data set; Based on the YOLOv3 code framework, load the pre-trained weights, use the perimeter intrusion image dataset to train the model, and obtain the initial intrusion monitoring model; Exporting an initial intrusion monitoring model, converting it into a node deployment format, optimizing the initial intrusion monitoring model based on the node deployment format, and obtaining a target intrusion monitoring model; The target intrusion monitoring model is deployed in the processing center of multiple terminal monitoring nodes, and is connected to the camera unit and the video recording unit through a wired communication.

4. The method according to claim 3, characterized in that After transmitting the first sub-region status result and the first sub-region abnormal video stream to the first processing center in the first terminal monitoring node, the method includes: Extracting the first sub-region real-time image corresponding to the first sub-region real-time state from the first sub-region state result; Inputting the first sub-region real-time image into the target intrusion monitoring model to obtain a first intrusion subject set, wherein different intrusion subjects in the first intrusion subject set have a subject unique identifier and a starting coordinate point, and the starting coordinate point has a coordinate point time identifier; Add the first intrusion subject set to the first terminal monitoring result.

5. The method according to claim 4, characterized in that After obtaining the first intrusion subject set, the method includes: Performing uniformly spaced image frame extraction on the abnormal video stream of the first sub-region to obtain a first abnormal image frame set, wherein each abnormal image frame in the first abnormal image frame set carries an image frame time mark; Based on the image frame time identifier, the abnormal image frames in the first abnormal image frame set are sequentially input into the target intrusion monitoring model in chronological order, and the first tracking monitoring results are sequentially output; The first tracking and monitoring result includes movement coordinate points of different intrusion entities, and the movement coordinate points are provided with coordinate point time identifiers, and the coordinate point time identifiers are consistent with the image frame time identifiers.

6. The method according to claim 5, characterized in that After obtaining the first tracking and monitoring result, the method includes: Matching the first intrusion subject set with the same intrusion subject in the first tracking and monitoring result based on the subject unique identifier, and obtaining an intrusion coordinate point set corresponding to the same intrusion subject; Based on the intrusion coordinate point set, taking the same intrusion subject as the fitting subject, taking the starting coordinate point as the intrusion starting point, taking the moving coordinate point as the intrusion moving point, performing curve fitting according to the coordinate point time mark, and acquiring a first intrusion route set; Add the first intrusion route set to the first terminal monitoring result.

7. The method according to claim 1, characterized in that The method further comprises: The first terminal monitoring node also includes a first infrared detection unit; When there is an abnormality in the status result of the first sub-area and the real-time light intensity of the first target sub-area is lower than the preset light intensity, the first infrared recording unit in the first terminal monitoring node is activated to record the first target sub-area and obtain the abnormal video stream of the first sub-area.

8. High-speed railway perimeter intrusion risk monitoring system based on multi-target recognition, characterized by: The method for monitoring the risk of high-speed railway perimeter intrusion based on multi-target identification according to any one of claims 1 to 7 is used to implement the method, the system comprises a distributed high-speed railway perimeter monitoring network, a wireless ad hoc network, and a railway control center, the distributed high-speed railway perimeter monitoring network comprises a plurality of terminal monitoring nodes, the terminal monitoring nodes comprise a camera unit, a video unit, and a processing center, the plurality of terminal monitoring nodes are connected to the railway control center through the wireless ad hoc network, and the system comprises: A monitoring network deployment module, the monitoring network deployment module is used to deploy the distributed high-speed railway perimeter monitoring network in the target high-speed railway perimeter area, and set the initial monitoring state of the target high-speed railway perimeter area for the distributed high-speed railway perimeter monitoring network; A real-time status acquisition module, the real-time status acquisition module is used to photograph a first target sub-area in the target high-speed railway perimeter area through a first camera unit in a first terminal monitoring node in the distributed high-speed railway perimeter monitoring network to acquire a real-time status of the first sub-area; an initial state calling module, the initial state calling module being used to call a first initial monitoring state of the first target sub-area in the initial monitoring state based on the real-time state of the first sub-area; A region status result module, the region status result module is used to compare the first sub-region real-time status with the first initial monitoring status to obtain a first sub-region status result; An abnormal video stream module, wherein when an abnormality occurs in the first sub-region status result, the abnormal video stream module is used to activate the first recording unit in the first terminal monitoring node, record the first target sub-region, and obtain an abnormal video stream of the first sub-region; A terminal monitoring result module, the terminal monitoring result module is used to transmit the first sub-region status result and the first sub-region abnormal video stream to the first processing center in the first terminal monitoring node, and analyze the first sub-region status result and the first sub-region abnormal video stream based on the first processing center to obtain a first terminal monitoring result, wherein the first terminal monitoring result includes a first intrusion subject set, a first intrusion route set, and a first intrusion risk level set; A monitoring visualization module is used to transmit the monitoring results of the first terminal to the railway control center in real time through the wireless ad hoc network, and the railway control center performs a visual display of the monitoring results of the first terminal.

Images