Railway perimeter and foreign matter intelligent monitoring and early warning method

Abstract

This invention provides an intelligent monitoring and early warning method for railway perimeter and foreign objects, belonging to the field of railway safety protection technology. The method includes: acquiring point cloud data of the railway's surrounding environment; making event judgments based on the point cloud data to generate corresponding tracking and pointing information; collecting corresponding video data based on the tracking and pointing information; using an artificial intelligence model to monitor perimeter intrusion and foreign object encroachment targets based on the video data; making fusion decisions based on the monitoring results of perimeter intrusion and foreign object encroachment; selecting whether to trigger a classification alarm based on the fusion decision results; if no classification alarm is triggered, continuing radar detection; and if a classification alarm is triggered, issuing a classification alarm. This method can provide early warning of the hazards of natural disasters to railway operations, promptly identify perimeter intrusion behavior, and issue warnings to management departments and train drivers.

Description

Technical Field

[0001] This invention relates to the field of railway safety protection technology, and more specifically, to a method for intelligent monitoring and early warning of railway perimeter and foreign objects. Background Technology

[0002] During the flood season, natural disasters such as landslides, mudslides, and rockfalls occur frequently, seriously threatening railway safety. At the same time, the unintentional or intentional trespassing of people and animals along the railway line is a persistent problem, potentially causing accidents. Traditional monitoring methods, such as manual inspections or single video surveillance, suffer from low efficiency, numerous blind spots, and significant susceptibility to environmental influences, making it difficult to achieve all-weather, high-precision real-time monitoring and early warning. Summary of the Invention

[0003] In view of this, the present invention proposes an intelligent monitoring and early warning method for railway perimeter and foreign objects to solve the problems existing in the prior art. The present invention provides an intelligent monitoring and early warning method that can predict the harm of natural disasters to railway operations, promptly identify perimeter intrusion behavior, and issue early warnings to management departments and train drivers, thereby ensuring the safety of passengers and railway facilities.

[0004] To achieve the above objectives, this invention proposes an intelligent monitoring and early warning method for railway perimeter and foreign objects, comprising: The system acquires point cloud data of the railway's surrounding environment, makes event judgments based on the point cloud data, and generates corresponding tracking direction information. Based on the tracking direction information, it collects corresponding video data and uses an artificial intelligence model to monitor targets such as perimeter intrusion and foreign object encroachment based on the video data. It then makes a fusion decision based on the monitoring results of perimeter intrusion and foreign object encroachment, and selects whether to trigger a classification alarm based on the fusion decision results. If no classification alarm is triggered, radar detection continues. If a classification alarm is triggered, a classification alarm is issued.

[0005] Optionally, the fusion decision includes judging the perimeter intrusion monitoring information and judging the foreign object intrusion monitoring results based on the three-dimensional information in the point cloud information.

[0006] Optionally, the process of judging perimeter intrusion monitoring information includes: In the perimeter intrusion monitoring information, determine whether the monitoring result is a human intrusion. If so, obtain and extract the monitoring personnel's work plan for verification to determine whether there is a relevant work plan. If not, issue an alarm. At the same time, if the monitoring result is an animal intrusion or an object intrusion, issue an alarm directly.

[0007] Optionally, during the process of judging the monitoring results of foreign object intrusion: In the monitoring results of foreign object encroachment, it is determined whether there are falling rocks, landslide objects, or abnormal roadbed objects. Based on the corresponding point cloud data, the relevant three-dimensional shape is judged. Based on the three-dimensional shape and the monitoring results of foreign object encroachment, it is determined whether a geological disaster phenomenon has occurred. If a geological disaster occurs, an alarm is issued directly; otherwise, it is recorded.

[0008] Optionally, the process based on the monitoring results of three-dimensional shape and foreign object intrusion limit includes: The suspected foreign object area identified in the video data is mapped to the radar coordinate system, and the corresponding point cloud cluster is extracted; the three-dimensional features of the point cloud cluster are calculated, and the three-dimensional features include one or more of the following: the size of the outer cuboid, volume, surface roughness, and reflection intensity statistics; the three-dimensional features are compared with a preset alarm threshold to comprehensively determine whether a foreign object intrusion event has been constituted and the event level.

[0009] Optionally, the artificial intelligence model adopts the YOLO algorithm model.

[0010] Optionally, after triggering a categorized alarm and performing the categorized alarm, the following may also be included: The system integrates and manages all alarm records for perimeter intrusion and foreign object encroachment, enabling the querying, processing, and statistics of early warning records; and provides real-time situation display and centralized traceability of early warning records.

[0011] On the other hand, the present invention provides an intelligent monitoring and early warning system for railway perimeter and foreign objects, used to perform the above-described method.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention utilizes the fusion monitoring of millimeter-wave radar and high-definition cameras, enabling the system to operate 24 / 7. It overcomes the effects of severe weather and insufficient lighting, significantly improving the detection rate and accuracy of targets such as falling rocks, landslides, and personnel intrusions, effectively reducing false alarms and missed alarms. Secondly, the deployment of edge computing terminals enables localized real-time data processing, greatly shortening the delay from detection to alarm, meeting the high real-time requirements of railway safety, and reducing the transmission pressure of massive amounts of video data. Thirdly, the system possesses intelligent decision-making capabilities, able to verify personnel identities in conjunction with work plans, and accurately analyze foreign object characteristics using 3D point cloud data, distinguishing between real threats and irrelevant interference, making alarms more targeted. Finally, the centralized management function of the cloud platform enables the integration, visualization, and multi-terminal linkage of monitoring data across the entire line, providing railway dispatching and maintenance personnel with convenient monitoring methods and decision support, thereby comprehensively improving the intelligence level of railway perimeter protection and foreign object intrusion warning. Attached Figure Description

[0013] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings: Figure 1 This is a flowchart illustrating the intelligent monitoring and early warning process for perimeter and foreign objects in an embodiment of the present invention. Figure 2 This is a diagram of the intelligent monitoring and early warning architecture for perimeter and foreign objects in an embodiment of the present invention; Figure 3 This is a system architecture diagram in an embodiment of the present invention; Figure 4 This is a diagram of the monitoring and early warning interface of the cloud platform in an embodiment of the present invention. Detailed Implementation

[0014] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0015] This embodiment proposes an intelligent monitoring and early warning method for railway perimeter and foreign objects, such as... Figure 1-3 As shown, it includes: The system detects point cloud data in the surrounding environment using millimeter-wave radar, and performs event judgment based on the point cloud data in the edge computing terminal to generate corresponding tracking and pointing information. Based on the tracking direction information, video image data in the corresponding direction is collected by a high-definition camera. In the edge computing terminal, the set artificial intelligence model performs target monitoring based on the video image data. The monitoring results of perimeter intrusion and foreign object intrusion are used to judge the perimeter intrusion monitoring information and combine the three-dimensional information in the point cloud information to judge the foreign object intrusion monitoring results for fusion decision. In the fusion decision, it is selected whether to trigger a classification alarm. If no classification alarm is triggered, radar detection continues. If a classification alarm is triggered, a classification alarm is triggered. The alarms, along with the corresponding monitoring results and video data, are then transmitted to the cloud platform via a wireless network for remote monitoring and alarm processing.

[0016] In the process of judging perimeter intrusion monitoring information, the first step is to determine the monitoring results. If the monitoring result is personnel intrusion, the work plan is extracted from the security protection system for verification to determine whether there is a relevant work plan. Feedback is then sent to the security protection system of the workers, and the relevant results are pushed. If the work plan is not in the work plan, an alarm is triggered. At the same time, if the monitoring result is animal intrusion or object intrusion, an alarm is triggered directly. In the process of judging the monitoring results of foreign object intrusion, the first step is to determine the type of monitoring target of the foreign object intrusion, whether it is a rockfall, landslide object or an abnormal roadbed object; then, the corresponding three-dimensional shape is judged by combining the radar point cloud data. Based on the three-dimensional shape and the relevant monitoring target type of foreign object intrusion, it is determined whether a geological disaster phenomenon has occurred. If a geological disaster occurs, an alarm is issued directly; otherwise, it is recorded.

[0017] The monitoring range for the aforementioned perimeter intrusion and foreign object encroachment is determined based on electronic fence technology.

[0018] By installing lidar and high-definition cameras, the system automatically identifies perimeter intrusion and foreign object encroachment incidents on-site and notifies the cloud platform to intervene and handle them. It also records, displays, alerts, and compiles statistics on perimeter intrusion and foreign object encroachment warnings.

[0019] The cloud platform integrates and manages all perimeter intrusion and foreign object intrusion warning records, allowing users to query all warning records here. The perimeter intrusion and foreign object intrusion monitoring equipment is a radar-visual integrated machine that records and manages the number of devices along the perimeter. Perimeter intrusion warnings are recorded in image format.

[0020] The intelligent monitoring and early warning method for railway perimeter and foreign objects provided by this invention relies on a system architecture consisting of front-end detection equipment and a cloud platform. The front-end detection equipment and the cloud platform communicate with each other via a wireless network. The front-end detection equipment includes a camera, a millimeter-wave radar, and an edge computing terminal. The aforementioned high-definition camera and millimeter-wave radar are mainly used to provide monitoring data to the edge computing terminal. The edge computing terminal processes the monitoring data, performs perimeter intrusion and foreign object intrusion monitoring and decision-making alarms based on the relevant monitoring data, and transmits the alarm results and monitoring data to the cloud platform. The cloud platform issues alarms and records the alarm results and monitoring data.

[0021] The above technical solution is described in detail: Landslides, mudslides, and rockfalls occur frequently during the flood season, seriously threatening railway safety. Constructing a railway foreign object intrusion monitoring and early warning system is essential to predict the potential harm of natural disasters to railway safety, issue early warnings, and notify trains in operation to take effective measures to ensure the safety of drivers and passengers. This invention proposes to use video surveillance as the primary monitoring method, employing millimeter-wave radar, high-definition cameras, and artificial intelligence technology to monitor intrusions by personnel and animals along the railway line. Based on the monitoring results, timely alerts will be sent to railway management departments and nearby train drivers to prevent accidents caused by intrusions by personnel and animals along the railway line.

[0022] like Figure 2-3 As shown, the system architecture for perimeter intrusion monitoring and early warning mainly includes front-end detection points and a cloud platform. The front-end detection points mainly consist of devices such as cameras, millimeter-wave radar, and edge computing terminals. The edge computing terminals intelligently analyze the collected video and millimeter-wave detection data to identify and analyze people, animals, and trains, and transmit the alarm data and the current image to the cloud platform via a 4G wireless network.

[0023] The perimeter intrusion monitoring and alarm system uses electronic fence technology to delineate protected areas along the railway line, collects video image data within a 1km monitoring section in real time, and classifies and detects intrusions by personnel, animals, and objects using edge computing based on built-in artificial intelligence algorithms. Simultaneously, the early warning system is linked to the operational safety protection system. Within the designated maintenance window, when personnel intrusion alarm information is detected, it promptly verifies the information in the railway operation plan. If the alarm information is confirmed to be within the operation plan, an early warning message is pushed to the relevant systems or terminals of the personnel involved, avoiding false alarms from personnel tagged within the operation plan. A maintenance window refers to a period of time reserved by the railway department for construction and maintenance work, during which train lines are not marked on the train timetable.

[0024] This invention aims to use video surveillance as the primary monitoring method, employing millimeter-wave radar, high-definition cameras, and artificial intelligence technology to monitor perimeter intrusion behaviors such as those by people and animals along railway lines. Based on the monitoring results, timely alerts will be sent to railway management departments and nearby train drivers to prevent accidents caused by intrusions by people and animals along railway lines.

[0025] The integrated radar-view system automatically identifies and analyzes real-time monitoring footage, uses artificial intelligence algorithms to identify unauthorized intrusions into the perimeter, and records, displays, alerts, and compiles statistics on perimeter intrusion warnings.

[0026] The aforementioned system architecture can also serve as a monitoring and early warning system for foreign object intrusion.

[0027] The Foreign Object Intrusion Monitoring and Alarm System provides monitoring and early warning for foreign object intrusion caused by landslides, mudslides, debris flows, and roadbed subsidence in areas prone to geological disasters. The foreign object intrusion function includes: 1. Automatic Monitoring: The system can automatically and continuously monitor the monitoring range. After identifying obstacles, it automatically analyzes and judges information such as the size and location of the obstacles, and automatically issues an early warning. 2. Alarm Notification: The system supports multiple concurrent alarms, including local speaker alarms, alarm light alarms, SMS alarms, and monitoring terminal alarms. 3. Real-time Monitoring: The control center supports the display of images from front-end monitoring points through the platform software client. It supports full-function remote control of front-end network cameras, including rotation, lens zoom, etc. 4. Video System Linkage: The system has a video linkage function. When there is no alarm, the video system periodically patrols the monitored area, forming patrol recordings; when the radar generates an alarm event, the video system dispatches appropriate cameras to track and locate the target, generating alarm recordings. The monitoring center platform software can display the on-site situation in a video pop-up window, allowing on-duty personnel to manually confirm the alarm.

[0028] For the aforementioned artificial intelligence model, the YOLO algorithm is used, and the YOLOv5 algorithm model can be used. The YOLOv5 algorithm model is trained and tested to generate monitoring results that can detect perimeter intrusion and foreign object encroachment.

[0029] In edge computing terminals, the YOLO series models are used as the core artificial intelligence algorithm to achieve real-time target monitoring of video image data.

[0030] The YOLO model works by dividing the input image into an S×S grid, with each grid cell responsible for detecting targets whose center point falls within that cell. For each grid cell, the model predicts B bounding boxes and their corresponding confidence scores, while also predicting the probabilities of C categories. The confidence scores reflect the degree of confidence that the bounding box contains the target and the degree of overlap between the predicted and ground truth boxes. Finally, redundant duplicate detection boxes are removed using a non-maximum suppression algorithm to obtain the final detection results.

[0031] For perimeter intrusion detection scenarios, the YOLO model is trained on a specialized dataset and can accurately identify intruding targets such as people and animals along railway lines. The training dataset contains a large number of labeled railway scene images, covering human and animal samples under different lighting conditions, poses, and distances, as well as interference samples under various complex backgrounds, ensuring that the model has strong generalization ability and anti-interference ability.

[0032] In actual operation, the edge computing terminal receives video streams captured by high-definition cameras and inputs each frame of the image into the trained YOLO model. The model performs forward inference calculations on the images and outputs the bounding box positions and category information of all detected targets in the image. For targets identified as "personnel," the system records their location coordinates, confidence scores, and timestamps to form personnel intrusion detection results; for targets identified as "animals," the same relevant information is recorded to form animal intrusion detection results.

[0033] Foreign object intrusion monitoring primarily targets static or dynamic obstacles such as rockfalls, landslides, and roadbed anomalies. The YOLO model is also specially trained to identify foreign object targets of different shapes and sizes. The training dataset contains a large number of images of geological hazards such as rockfalls, landslides, and collapses, as well as background images of normal railway track beds and slopes, enabling the model to accurately distinguish abnormal objects from the normal environment.

[0034] Since foreign object intrusion targets often have irregular three-dimensional shapes, relying solely on two-dimensional image recognition may have limitations. Therefore, the output of the YOLO model is fused with three-dimensional point cloud data acquired by millimeter-wave radar. The model first identifies regions suspected to be foreign objects from video images, outputting target bounding boxes and preliminary category judgments. Subsequently, based on the image region corresponding to the bounding box, the system extracts the three-dimensional spatial information of that region from the radar point cloud data, including the target's height, width, depth, and estimated volume.

[0035] By fusing image recognition and point cloud data, the system can more accurately classify and identify foreign objects. For example, for a target initially identified as "falling rock" by the YOLO model, the system combines point cloud data to analyze whether its three-dimensional shape has typical characteristics of rock, whether its volume exceeds a preset threshold, and whether its location intrudes into the railway clearance, thereby comprehensively determining whether it constitutes a genuine foreign object intrusion event. For landslides and roadbed anomalies, the system focuses on their area range, height changes, and comparison with historical data to determine whether an alarm threshold has been reached.

[0036] The final output of the YOLO model includes the class label, bounding box coordinates, and confidence score for each detected target. Edge computing terminals post-process these outputs to generate structured monitoring results data.

[0037] For perimeter intrusion monitoring, the monitoring results include intrusion type (person or animal), intrusion time, intrusion location (image coordinates and corresponding actual geographic coordinates), target movement direction, confidence score, and associated original image frames. This information is passed to the decision-making module for subsequent work plan verification and alarm decision-making.

[0038] For foreign object intrusion monitoring, the monitoring results include the type of foreign object (rockfall, landslide, roadbed anomaly, etc.), detection time, location coordinates, estimated 3D dimensions, estimated volume, confidence score, and associated original images and point cloud data. This information is passed to the fusion decision module, which combines the point cloud 3D information to make a final judgment.

[0039] Deploying the YOLO model on edge computing terminals, located close to the data acquisition source, enables millisecond-level real-time inference, meeting the low-latency requirements of railway safety monitoring. Secondly, large amounts of raw video data are processed locally, with only alarm information and key images uploaded, significantly reducing the pressure on wireless network bandwidth. Furthermore, edge deployment enhances system reliability; even during network outages, front-end devices can still independently perform monitoring and alarm functions, and data is synchronized to the cloud only after network recovery.

[0040] After generating the corresponding monitoring results, relevant fusion decisions are made, including the following: In the edge computing terminal, the YOLOv5 model outputs the category confidence score and bounding box coordinates for each detected target. The system further categorizes the monitoring results into four classes based on the category label: "personnel," "animals," "large objects" (such as construction machinery), and "other objects." For each target, a unique tracking ID, timestamp (millisecond level), image coordinates, and WGS84 latitude and longitude coordinates and railway mileage obtained through camera calibration parameters and electronic map conversion are attached.

[0041] When the monitoring result identifies a person, the work plan verification process is immediately initiated. The verification request is sent to the cloud platform's security protection system via 4G / 5G network. Request parameters include: timestamp, location, camera device ID, target tracking ID, and target screenshot. Upon receiving the request, the cloud platform first queries the active work plans within the last hour. If no match is found, it retrieves the detailed information of the work plan from the MySQL sharded database. The matching algorithm employs a dual spatial-temporal index: temporally, a B+ tree index is used to quickly locate the record covering the request time window of the plan; spatially, an R-tree index is used to calculate the minimum distance between the requested location and the planned electronic fence polygon. If the distance is less than a set threshold, it is considered a location match. For plans with spatiotemporal matching, further facial recognition comparison is performed: if the target screenshot's clarity meets the requirements, features are extracted from the cloud facial feature library and compared 1:N with the registered photos of the personnel in the plan. A cosine similarity exceeding 0.6 is considered a personnel match. Verification results are categorized into three types: complete match (time + location + personnel all match), partial match (time + location only match but the personnel are not on the list), and no match (no plan match). The results are returned to the edge terminal and simultaneously written to the relevant records on the cloud platform. For complete matches, the system pushes a notification to the handheld terminal of the corresponding construction supervisor via the MQTT protocol, indicating that the worker has been identified, including the worker's name, location, and time. For partial matches, an intrusion warning for unknown personnel is sent to the supervisor, prompting verification.

[0042] Animal and object intrusions trigger alarms directly without verification. However, multi-frame verification is required before an alarm can be triggered: the edge computing unit maintains a Kalman filter tracker that performs cross-frame association for each target. Only when the same tracking ID is detected in five consecutive frames with an average confidence level greater than 0.7 is it confirmed as a valid intrusion. The tracker uses target position and speed information to predict the area where it will appear in the next frame, reducing missed detections. After a valid intrusion is triggered, the system immediately generates an alarm event, uploads it to the cloud via wireless network, and simultaneously activates the on-site audible and visual alarms. The alarm event includes: event ID, type, time, precise location, target screenshot, confidence sequence, and associated video clips. After receiving the alarm, the cloud platform pushes it to relevant dispatch consoles, station duty officer terminals, and nearby locomotive onboard terminals according to preset rules.

[0043] The YOLOv5s model for foreign object intrusion monitoring adds a multi-scale feature fusion layer to the standard detection head to improve the perception of large targets (such as landslides). The model training set contains 100,000 labeled images, covering rockfalls, landslides, roadbed anomalies, and other foreign objects under different lighting, weather, and seasons. In addition to outputting class probabilities, the model also outputs a precise segmentation mask of the target in the image, facilitating subsequent 3D mapping.

[0044] When the model detects a suspected foreign object, the system first determines the region of interest (ROI) in the image based on the target bounding box and segmentation mask. Using a pre-calibrated camera-radar joint extrinsic parameter matrix, this region is mapped to the radar coordinate system, forming a three-dimensional cone space. The system extracts the point cloud within the cone from the millimeter-wave radar and performs preprocessing: removing ground points and clustering to obtain target point cloud clusters. For each point cloud cluster, the following three-dimensional features are calculated: Outer cuboid dimensions: Principal component analysis is used to obtain the three principal axes of the point cloud, and the projected length is calculated to obtain the length, width, and height, with an accuracy of up to 0.1m. Volume: The convex hull volume of the point cloud cluster is calculated using a convex hull algorithm; for sparse point clouds, a voxelization method is used to accumulate occupied voxels. Surface roughness: The entropy of the point cloud normal vector distribution is calculated to distinguish between rocks (high entropy) and artificial objects (low entropy). Reflection intensity statistics: Millimeter-wave radar reflection intensity can help determine the material; rock reflection intensity is usually higher than soil.

[0045] Based on the above three-dimensional features, relevant alarms are issued: Rockfall alarm: It must simultaneously meet the following conditions: (1) The image is classified as rockfall and the confidence level is >0.8; (2) Any dimension of the outer cuboid is >0.3m; (3) The volume is >0.05m. 3 (4) The convex hull shape is irregularly blocky (major axis length ratio > 2.5 and surface roughness > 0.7); (5) The target center point is located within 2.5m on both sides of the track centerline or in the area above the slope where it may roll off. In addition, if the target is detected to move at a speed > 0.2m / s and point towards the track in 3 consecutive frames, the emergency alarm level is upgraded to Level 1.

[0046] Landslide alarm: must meet the following conditions: (1) the image is classified as a landslide and the confidence level is >0.75; (2) the base area of ​​the outer cuboid is >10m². 2 (3) The point cloud shows a significant elevation difference >1m; (4) Compared with historical point clouds, the volume change rate is >20% or the newly appearing area is >5m. 2 For landslides, the system also calculates the slip vector and predicts the potential impact range.

[0047] Subgrade anomaly alarm: based on track section template matching. The system pre-establishes a standard track section point cloud template. During real-time monitoring, the current track bed area point cloud is registered with the template using ICP, and the vertical deviation of each sleeper position is calculated. When the deviation of 5 consecutive sleepers exceeds 5cm and the cumulative length is >10m, a subgrade anomaly alarm is triggered, and the anomaly location and deviation amount are marked.

[0048] All foreign object detection is verified using multiple frames, but for dynamic targets such as rolling stones, a shorter confirmation frame count is used to gain more warning time. After successful verification, the system uploads the alarm information along with the associated point cloud data and video clips to the cloud. The cloud then performs 3D reconstruction of the point cloud data to create a visual model for remote viewing by dispatchers.

[0049] The rule engine embedded in the edge computing terminal receives real-time judgment results from two sub-modules: perimeter intrusion and foreign object intrusion. Each result contains a priority field, assigned a value by the system according to predefined rules: Priority 0 (highest): SOS proactive trigger, sudden disasters when a train approaches. Priority 1: Geological disaster alarm, including rockfalls, landslides, and roadbed anomalies. Priority 2: Personnel intrusion alarm, including unverified personnel. Priority 3: Animal / object intrusion alarm.

[0050] Within the cloud platform, the homepage monitoring interface centrally displays the real-time status of perimeter intrusions, facilitating rapid access to overall information for dispatchers and station staff. The homepage is divided into two parts: data filtering criteria and data display. The data display includes five core modules: perimeter intrusion warning status, perimeter intrusion warning ranking, warning reason statistics, warning scrolling, and warning status. For data filtering, dispatchers can filter by the entire line, by route, by station, and by time period. Clicking "Entire Line Statistics" displays data for the entire line, while clicking on a specific route or station displays data for that corresponding range. The interface automatically refreshes when the time period changes. When a perimeter intrusion warning occurs, station users can click "Process" to open a processing window. Within this window, clicking on a photo enlarges it to view the video. After confirmation, the warning is marked as a false alarm, a person, or an animal, depending on the actual situation. Once marked, the warning status changes to "Processed," and the processing result is automatically synchronized to the dispatcher. The data fields displayed in the warning ranking include ranking, station, and number of warnings. The station side also displays the perimeter type and mileage. The warning scrolling display shows the station, perimeter protection type, perimeter protection mileage, camera number, and number of warnings. The warning status displays the warning time, warning type, warning station, perimeter protection type, perimeter protection mileage, perimeter protection line, and device number. For example... Figure 4 As shown, the relevant interface for personnel monitoring is presented.

[0051] In the device management module, the system performs full lifecycle management of perimeter intrusion devices, including basic information maintenance, viewing of warning records, adding devices, and batch operations. Device filtering criteria include device registration location, perimeter type, device model, device type, manufacturer name, device status, device entry time period, and device number. Device list fields include serial number, device registration location, perimeter type, mileage, device number, device type, manufacturer name, manufacturer number, area location, device status, warning record creator, creation time, and operation. When adding or modifying devices, the registration location, perimeter type, device type, device model, device number, manufacturer name, manufacturer number, row, and device latitude and longitude must be filled in. Warning record viewing can be filtered by perimeter type, mileage, row, and device number, and also supports filtering by personnel, animals, or time period. The warning list displays the warning time, warning type, and warning content, and users can click to view video surveillance. When handling an alert, clicking the "Handle" button will open a pop-up window that displays the perimeter type, mileage, row, device number, alert content, alert mileage, alert time, alert type, and alert details. It also provides options for false alarms, confirming personnel, and confirming animals, and allows you to view the monitoring data.

[0052] In the Early Warning Center module, the system centrally manages all perimeter intrusion warning records for easy querying and tracing. The Early Warning Center displays all perimeter intrusion warning messages under the account's permissions, supporting multi-dimensional filtering, including intruder type, warning status, protection type, warning location, warning time period, and mileage. The warning list includes the sequence number, intruder type, perimeter type, warning location, mileage, row number, warning status, warning time, and action. Clicking "Process" allows you to mark the warning as a false alarm, confirm the person, or confirm the animal. Clicking "Details" displays complete warning information, including warning mileage, warning type, warning time, and warning content.

[0053] On the cloud platform, foreign object intrusion monitoring and early warning are implemented. The homepage monitoring interface displays foreign object intrusion events in real time and provides a processing entry point. The homepage is divided into two parts: data filtering conditions and data display. The data display is further subdivided into five modules: foreign object intrusion early warning status, foreign object intrusion early warning ranking, early warning cause statistics, early warning scrolling, and early warning status. For data filtering, dispatch users can filter by the entire line, line, station, and time period, and refresh the data by clicking the corresponding range. Station users can filter by time period and view the list of passageway doors. In the early warning processing flow, station users can click the processing pop-up window to zoom in on the on-site image, mark it as a false alarm, stone, or branch according to the actual situation, and after processing, the status changes to "processed" and is synchronized to the dispatch terminal. The early warning ranking displays the ranking, station, and number of warnings; the station terminal also displays the foreign object type and mileage. The early warning scrolling displays the station, foreign object protection type, foreign object protection mileage, camera number, and number of warnings. The early warning status displays the early warning time, early warning type, warning station, foreign object protection type, foreign object protection mileage, foreign object protection line, and equipment number.

[0054] In the equipment management module, the system provides unified management of foreign object intrusion devices. Equipment filtering criteria include equipment registration location, foreign object protection type, equipment model, equipment type, manufacturer name, equipment status, equipment entry time period, and equipment number. The equipment list fields include serial number, equipment registration location, foreign object protection type, mileage, equipment number, equipment type, manufacturer name, manufacturer number, area location, equipment status, alert record creator, creation time, and operation. When adding or modifying equipment, the registration location, perimeter type, equipment type, equipment model, equipment number, manufacturer name, manufacturer number, row, and equipment latitude and longitude must be filled in. Alert records can be viewed by perimeter type, mileage, row, and equipment number, and also by personnel, animals, or time period. The alert list displays the alert time, alert type, and alert content, and monitoring data can be viewed.

[0055] In the Early Warning Center module, the system centrally manages all foreign object intrusion warning records. The Early Warning Center displays all foreign object intrusion warning messages under the account's permissions, supporting multi-dimensional filtering, including intruder type, warning status, protection type, warning location, warning time period, and mileage. The warning list includes the sequence number, intruder type, protection type, warning location, mileage, line number, warning status, warning time, and action. Clicking "Process" allows for marking false alarms, confirming stones, or confirming tree branches; clicking "Details" displays the warning mileage, warning type, warning time, and warning content.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for intelligent monitoring and early warning of railway perimeter and foreign objects, characterized in that, include: Acquire point cloud data of the railway's surrounding environment, make event judgments based on the point cloud data, and generate corresponding tracking and pointing information; Based on the tracking information, corresponding video data is collected. An artificial intelligence model is used to monitor perimeter intrusion and foreign object encroachment based on the video data. A fusion decision is made based on the monitoring results of perimeter intrusion and foreign object encroachment. Based on the fusion decision results, it is selected whether to trigger a classification alarm. If no classification alarm is triggered, radar detection continues. If a classification alarm is triggered, a classification alarm is issued.

2. The method according to claim 1, characterized in that, The fusion decision-making process includes judging the perimeter intrusion monitoring information and judging the foreign object intrusion monitoring results based on the three-dimensional information in the point cloud information.

3. The method according to claim 2, characterized in that, The process of judging perimeter intrusion monitoring information includes: In the perimeter intrusion monitoring information, determine whether the monitoring result is a human intrusion. If so, obtain and extract the monitoring personnel's work plan for verification to determine whether there is a relevant work plan. If not, issue an alarm. At the same time, if the monitoring result is an animal intrusion or an object intrusion, issue an alarm directly.

4. The method according to claim 2, characterized in that, In the process of judging the monitoring results of foreign object intrusion: In the monitoring results of foreign object encroachment, it is determined whether there are falling rocks, landslide objects, or abnormal roadbed objects. Based on the corresponding point cloud data, the relevant three-dimensional shape is judged. Based on the three-dimensional shape and the monitoring results of foreign object encroachment, it is determined whether a geological disaster phenomenon has occurred. If a geological disaster occurs, an alarm is issued directly; otherwise, it is recorded.

5. The method according to claim 1, characterized in that, The process based on the monitoring results of three-dimensional shape and foreign object intrusion limit includes: The suspected foreign object area identified in the video data is mapped to the radar coordinate system, and the corresponding point cloud cluster is extracted; the three-dimensional features of the point cloud cluster are calculated, and the three-dimensional features include one or more of the following: the size of the outer cuboid, volume, surface roughness, and reflection intensity statistics; the three-dimensional features are compared with a preset alarm threshold to comprehensively determine whether a foreign object intrusion event has been constituted and the event level.

6. The method according to claim 1, characterized in that, The artificial intelligence model uses the YOLO algorithm.

7. The method according to claim 1, characterized in that, After triggering a categorized alarm, the following steps are also included: The system integrates and manages all alarm records for perimeter intrusion and foreign object encroachment, enabling the querying, processing, and statistics of early warning records; and provides real-time situation display and centralized traceability of early warning records.

8. A railway perimeter and foreign object intelligent monitoring and early warning system, characterized in that, Used to perform the method described in any one of claims 1-7.

Images