Railway line patrol safety intelligent early warning system and method based on edge computer

Abstract

This invention discloses an intelligent early warning system and method for railway line patrol safety based on edge computing, involving intelligent safety early warning in the railway transportation field, enabling real-time controllable emergency measures to minimize the negative impact of incidents. The edge computing-based intelligent early warning system for railway line patrol safety includes: a cloud platform for an intelligent patrol system along the railway line, network infrastructure units for the intelligent patrol system along the railway line, on-site edge intelligent early warning devices and traffic control units, and IoT video sensing infrastructure units. This invention aims to deploy edge computers to continuously analyze 24-hour video streams from surveillance cameras at key locations along the railway line, process data in real-time, reduce latency, promptly identify abnormal events, take temporary emergency measures as quickly as possible, and rapidly report abnormal events, effectively shortening response time, significantly improving the efficiency of handling abnormal events in railway transportation, enhancing the accuracy of emergency measures, and ensuring railway line safety.

Description

Technical Field

[0001] This invention relates to the field of railway transportation and safety technology, and more specifically, to the infrastructure construction of an artificial intelligence technology application based on edge computing, focusing on the construction and technological promotion and application of an intelligent early warning system for railway line patrol safety. Background Technology

[0002] Currently, railway line patrols still rely primarily on manual inspections. This is particularly problematic in areas with harsh natural conditions, where manual inspections are unsuitable. Manual data collection and the detection of potential anomalies are extremely difficult, time-consuming, and inefficient, while also posing health and safety risks to patrol personnel. Protective fences and walls with anti-climbing devices are installed along railway lines and at key traffic locations according to standards to prevent and stop behaviors that endanger railway transport safety, effectively ensuring railway safety. While the construction of integrated video surveillance along railway lines allows for real-time video monitoring of key areas, this method requires back-end personnel to manually inspect large amounts of image data, resulting in low efficiency and limited coverage. Centralized video surveillance management heavily relies on networks, leading to significant latency, heavy central workload, poor scalability, untimely emergency response to abnormal events, and delayed incident handling, failing to effectively curb the development of accidents.

[0003] The development of information technology and intelligent railway transportation technology has brought new opportunities for improving the safety level along railway lines. With the promotion and construction of intelligent transportation, advanced technologies such as artificial intelligence and big data analysis are being applied more widely in the field of railway traffic control. The continuous rise and development of edge computing technology has provided railway traffic safety patrol systems with more powerful data processing and decision-making capabilities, greatly expanding the application scenarios and scope of railway traffic safety control. Based on edge computing devices and intelligent algorithms, and building upon the extensive deployment of integrated video surveillance along railway lines, traffic data can be collected and analyzed in real time and accurately, providing real-time decision-making and response, autonomously issuing traffic safety warnings, reducing accident rates, and improving railway safety. At this stage, the railway traffic safety patrol system possesses self-learning and adaptability capabilities, achieving more efficient and safer management of the railway traffic safety patrol system. Summary of the Invention

[0004] This invention enables real-time alarms for safety hazards and disasters along railway lines, improving the level of technical prevention; it fully utilizes the system to reduce the workload and intensity of on-site manual inspections, improving inspection efficiency and quality; it uses intelligent analysis algorithms to reduce the workload of manual video inspections and compensate for loopholes caused by human fatigue and negligence; it standardizes railway inspection management requirements, and only by making good use of artificial intelligence technology prevention measures can the standardization of railway line inspection operations be implemented; it improves the level of hazard prevention in key high-risk scenarios (tunnel entrances, open tunnels, aqueducts, bridges, stations, etc.), ensuring the safety of railway operations and the safety of people's lives and property.

[0005] The specific technical solution for implementing this invention is as follows: Step 1: This invention provides an intelligent early warning system for railway line patrol safety based on edge computing. The system includes: a railway line intelligent patrol system cloud platform S1, a railway line intelligent patrol system network infrastructure unit S2, an on-site edge intelligent early warning device and traffic control unit S3, and an Internet of Things (IoT) sensing facility unit S4; (Reference) Figure 1 ; The intelligent railway patrol system cloud platform S1 includes the China Railway Administration central cloud platform S10 and regional bureau cloud platforms S11. The central cloud platform S10 manages all regional cloud platforms, monitors the safety status of railway lines across the entire railway network, analyzes the impact of abnormal events on the overall railway network operation, and makes timely decisions to allocate resources, minimizing the negative impacts of various anomalies. The regional bureau cloud platform S11 manages all intelligent early warning devices within its region, monitors railway line patrols within its region, analyzes the causes and countermeasures of abnormal events within its region, and takes necessary emergency measures. (Reference) Figure 4 , Figure 5 ; The intelligent patrol system network infrastructure unit S2 along the railway line includes the China Railway Communication private network and the public networks of the three major telecom operators along the railway line, as well as the station local area networks built on these two major network communication facilities, and their power supply guarantee facilities S20, etc., and all facilities and methods to ensure network security and stable operation; (Reference) Figure 3 ; The on-site edge intelligent early warning device and traffic control unit S3 includes all edge computers installed at key monitoring locations, alarm systems requiring real-time control, other automated emergency equipment, and power supply protection equipment for the aforementioned facilities; this unit is the core facility for achieving the purpose of this invention and a key equipment unit for ensuring that various abnormal events are handled correctly and effectively in real time; Reference Figure 2 ; The IoT sensing facility unit S4 includes all IoT video sensing facility units along the railway line, as well as other sensing equipment required for line patrol, such as radio frequency sensors for precise control of carriage position sensing; (Reference) Figure 2 .

[0006] Step 2: The IoT video sensing facility unit S4 in Step 1 is characterized by including a network video camera S41 and a power supply and backup power system S42 to ensure its normal operation. Since railway lines typically traverse remote areas with complex geological and climatic conditions, the power supply unit must consider backup power systems such as photovoltaic and battery power supplies to maintain power for a period of time in case of problems with the normal power supply facilities, ensuring that the video cameras can upload video streams of a certain duration for event analysis. (See reference...) Figure 2 .

[0007] Step 3: The power management unit S30 of the field edge intelligent early warning device and traffic control unit S3 in Step 1, like the video camera in Step 2, consumes more power due to the similar environmental characteristics. Therefore, a more powerful power management unit is needed to ensure that the edge computer can maintain a certain working time, analyze and judge abnormal events in various video streams, and accurately and timely upload them to the regional cloud platform; (Reference) Figure 2 ; The field edge computer S31 must have the following functions: field IPC video device management (S311), task orchestration management (S312), algorithm repository management (S313), network settings management (S315), system settings management (S316), alarm upload and field alarm management, and anomaly policy management, etc.; (See reference) Figure 2 ; The S311 field IPC video device management function is characterized by a field edge computer that enables real-time video browsing and camera management for viewing various devices on-site. The video list allows for browsing, viewing, editing, and deleting of camera videos, and also allows for the addition and management of other IPC video devices. (See reference...) Figure 2 ; The task orchestration and management function S312 includes task orchestration and task recording. Devices are orchestrated and managed in the task center. When adding a task, steps such as "task name," "recognition interval," "camera selection," "algorithm loading," and "filtering mode" are completed. After selecting the camera, the required algorithm is loaded, and the filtering mode is selected. During configuration, algorithms can be freely plugged in and out. The bounding box of the target object can be selected as needed: "completely within the ROI," "center point within the ROI," or "overlapping with the ROI." Other values ​​will be filtered and not displayed in the task record. The task record records and traces the events in which the device algorithm takes effect; the record list corresponds to the task list. (Reference) Figure 2 , Figure 5 ; The algorithm repository management function S313 allows users to view authorized algorithms in the system's algorithm repository. Authorized algorithms are distributed from the regional cloud platform to the edge computing platform for authorization, followed by task management and other operations. (See reference...) Figure 5 ; The upload alarm and on-site alarm management function S314 triggers the edge computer platform to report abnormal events to the higher-level regional cloud platform if an abnormal event occurs in the video stream for various functions and purposes. Simultaneously, it uploads relevant video evidence, triggers on-site alarms to issue alarm signals, and initiates necessary temporary emergency measures. The on-site emergency event intelligent control and processing function S318 minimizes the impact of abnormal events as much as possible. (Reference) Figure 2 , Figure 5 ; The S315 network settings management tool allows you to configure or modify network settings, default gateway, IP address, subnet mask, and DNS. After making changes, click "OK" to save the configuration. (See reference...) Figure 2 ; The S316 system configuration function displays the device name, allows modification, and saves the configuration. The device time has already been synchronized, either via NTP (Network Time Protocol) or manually, before saving the configuration. (See reference...) Figure 2 ; The S317 exception policy management function includes rule management and rule recording. Rule management manages added tasks, including ID, rule name, rule template, rule status, and creation time. It allows viewing and deleting rules. Adding a rule includes information such as "Rule Name," "Task Source," "Rule Template," and "Remarks." The task source is selected based on the task name created in the task center. The rule template allows selection of target dwell timeout, target quantity timeout threshold, and target long-term loss. After adding remarks, clicking "Next" selects the ROI. The ROI selection box is based on the ROI number created in the task center. After selecting the event name, clicking "Next" sets the ROI. The ROI settings are based on the rule template, which considers both time and quantity. The target dwell timeout and target long-term loss options allow for custom timeout durations (in seconds), and the target quantity timeout threshold allows for custom recognition counts (in units). Rule recording records the events that take effect under rule management. Rule records include ID, rule name, task name, camera name, rule template, occurrence time, and thumbnail. They can be viewed and traced back by querying the rule name, task name, camera name, and occurrence time. (Reference) Figure 2 .

[0008] Step 4: The network facility unit S2 of the intelligent patrol system along the railway line mentioned in Step 1 has an underlying facility consisting of local area networks (LANs) composed of edge computers at key locations along the railway line, several video cameras (or other railway signal sensing sensors), and network devices (routers, industrial gateways, etc.). These facilities also need to provide a powerful power management unit (field network device power management) S20, similar to the environmental characteristics of the video cameras in Step 2. The power supply mode of these network devices can be a backup power supply such as a UPS power supply to ensure that the network infrastructure such as routers and gateways can maintain a certain working time and ensure that the edge computers can reliably transmit information to the cloud platform when abnormal events occur. The LAN can be composed of a combination of wired network and wireless WiFi network to meet the field networking needs of various heterogeneous devices. Several m LANs based on edge computers will aggregate information to the regional bureau cloud platform through the railway communication network. The railway communication network includes China Railcom's proprietary network system S21 and the public network system S22 of the three major operators (China Mobile, China Unicom, China Telecom, etc.). China Railcom's proprietary network system S21 includes the railway communication 4G / 5G wireless network S23 and the railway communication dedicated wired network S24. The wired network is a dedicated network for normal use. When the wired network fails due to natural disasters, damage to the main power supply line, or other reasons, the wireless network serves as a supplementary backup network, or the 4G / 5G wireless network can be used directly in places where it is inconvenient to lay cables. The public network system S22 of the three major operators (China Mobile, China Unicom, China Telecom, etc.) is a supplementary backup network for this railway dedicated line network, and also includes the 4G / 5G wireless network S263 and the wired network along the railway line S25. The infrastructure of the railway center network in the intelligent patrol system network facility unit S2 along the railway line is also based on the principle of using China Tietong's dedicated network as the main network and the networks of the three major telecom operators as supplementary networks; (Refer to...) Figure 3 .

[0009] Step 5: The S1 infrastructure of the intelligent patrol system cloud platform along the railway line mentioned in Step 1, on the one hand, establishes a dedicated cloud server of China Railcom to maintain the normal operation of the cloud platform, and on the other hand, other commercial servers can be used as backup means to ensure the reliable operation of the information link. The General Administration's central cloud platform S10 includes central control functions S101, intelligent algorithm training functions S102, security module functions S103, cloud computing functions S104, and intelligent analysis and decision-making functions S105; it monitors the status of all railway lines in the national railway system, collects information from n regional bureau cloud platforms, and references... Figure 4 ; The central control function S101 includes network control, security control, computing power control, data control, node control, and control of various scenarios, etc., automatically monitoring the status of railway lines and automatically optimizing control measures; The intelligent algorithm training function S102 is a reliable source of algorithms for various scenarios. Only the China Railway Administration possesses comprehensive direct data on various scenarios, boasting unparalleled big data conditions that no other company or social organization can match. This forms a solid foundation for optimizing and training models for various scenario algorithms. Currently, the urgently needed scenario algorithms are as follows: Intelligent early warning algorithm for crossing railings S1020; Intelligent early warning algorithm for geological disasters such as landslides and rockfalls at tunnel entrances S1021; Intelligent early warning algorithm for geological disasters such as flooding of railways S1022; Intelligent early warning algorithm for foreign objects on overhead power lines S1023; Intelligent algorithm for personnel or large animals encroaching on rail limits S1024; Intelligent early warning algorithm for smoke and fire alarms S1025; Intelligent early warning algorithm for platform personnel encroaching on rail limits S1026; Intelligent early warning algorithm for platform personnel falling S1027; Intelligent early warning algorithm for debris flow S1028; Intelligent early warning algorithm for monitoring crowd congestion at station exits S1029, etc. Security module S103 refers to firewalls, antivirus, intrusion detection, intrusion prevention, log auditing, network gateways, etc. As the main artery of international economic operation, the security of the railway platform is extremely important. Any input and output of information data must be isolated and processed by security modules to meet national security level requirements. The cloud computing function S104 is the computing power within the cloud platform designed to meet the needs of data analysis. The intelligent analysis and decision-making function S105 is a manifestation of the role of the cloud platform. All the data acquired is for the cloud platform, which acts as the brain, to perform rapid calculations on the server and provide correct decision-making references based on the correct rules and constraints of railway operation. These references are then pushed to the railway dispatching system S12 for train operation scheduling reference. The regional bureau cloud platform S11 includes a regional cloud algorithm warehouse function S111, a powerful regional center control function S112, a cloud-edge high-efficiency collaboration function S113, and a security module S114; it monitors the status of all edge computers within the regional bureau, obtains the status of railway lines within the region from the edge computers, and actively pushes it to the central cloud platform of the headquarters; (Reference) Figure 5 ; The algorithms in the regional cloud algorithm warehouse function S111 are mostly provided by the intelligent algorithm training function center S102 of the China Railway Corporation cloud platform, and a small part of the algorithms can come from artificial intelligence algorithm companies other than the China Academy of Railway Sciences. The powerful regional central control S112 optimizes the functional application distribution of edge computers within the region, rationally manages computing resources, arranges various algorithm function requirements in places where they are urgently needed, manages network node data flow, and ensures reasonable allocation of network resources. The cloud-edge high-efficiency collaboration function S113 refers to the integration of computing power, storage, network and other resources of edge computers into servers, helping users to quickly build edge computing cloud network environments, supporting the network and computing power needs of intelligent applications at low cost and high efficiency; regional cloud realizes unified orchestration, scheduling, configuration and management of complex network, computing power, data, application and other resources on the edge side, shielding the complexity of the edge side, and providing the edge side with AI, IoT, security and other capabilities support; Security module S114, like S103 of the central cloud platform, ensures that only edge computers authorized by the regional bureau platform can access the cloud platform, and isolates unauthorized devices from intruding into the regional cloud platform system.

[0010] Step 6: Most of the scene algorithms in the intelligent algorithm training function S102 described in Step 5 are artificial intelligence algorithms based on video streams. Their basic workflow diagrams are roughly the same; the process is roughly as follows: P000 The central cloud platform of the intelligent patrol system along the railway line provides various intelligent early warning algorithms approved by the General Administration; P100 The cloud platform of the railway regional bureau receives the intelligent early warning algorithms authorized by the General Administration platform; P110 The cloud platform of the railway regional bureau distributes various intelligent early warning algorithms to the edge computer units with different algorithm requirements within the region; P120 The edge computer units assign the intelligent early warning algorithms to various video stream sources with different application requirements. The video stream sources come from cameras installed at different application requirement locations; P130 Various video streams that meet the requirements of different algorithms are obtained from various cameras; P140 Does the intelligent algorithm identify the assigned abnormal situation from the associated video stream? If no abnormal event is identified, the process returns to P130 to continue working. If an abnormal event that meets the algorithm's trigger requirements is identified, P150 records the abnormal event on the local edge computer and initiates on-site emergency measures. P160 intelligently processes on-site warnings and alerts, reports the abnormal event to the cloud platform, and pushes relevant video streams. P170 continues to acquire various video streams from different cameras that meet different algorithm requirements. P180 checks if the intelligent algorithm has identified the assigned abnormal situation from the associated video streams. If it is still an abnormal situation, the process returns to P170 to continue working. If the abnormal event has been identified and processed, P190 deactivates the on-site alarm settings, allowing manual restoration of emergency measures to normal status. P200 sends a message to the cloud platform center that the abnormal event has been processed and the alarm has been deactivated, then returns to P130 to continue monitoring the railway line status. (Reference) Figure 6 .

[0011] Step 7, the S1020 intelligent hurdle crossing warning algorithm in the S102 intelligent algorithm training function mentioned in Step 5, is characterized by analyzing the collected hurdle crossing video data and generating a warning strategy based on the analysis results, including the following steps: Use a camera to capture video streams of objects crossing railings; A deep learning-based fence crossing event detection algorithm is used to analyze video data, and the existence of fence crossing events is determined based on the analysis results. If a fence crossing event is detected, an early warning strategy is generated based on the fence crossing event type and the location where the fence crossing event was detected; if no fence crossing event is detected, the fence crossing detection video stream is read again.

[0012] Step 8, the S1021 intelligent early warning algorithm for geological disasters such as landslides and rockfalls at tunnel entrances, which is part of the S102 intelligent algorithm training function mentioned in Step 5, is characterized by analyzing the collected video data of landslides and rockfalls at tunnel entrances and generating early warning strategies based on the analysis results, including the following steps: Using cameras to capture video streams of landslide and rockfall detection at tunnel entrances; A tunnel entrance landslide and rockfall event detection algorithm based on kinematics and the law of conservation of energy is used to analyze video data, and the existence of a tunnel entrance landslide and rockfall event is determined based on the analysis results. If a landslide and rockfall event occurs at the tunnel entrance, an early warning strategy is generated based on the type of landslide and rockfall event and the location where the event is detected. If no landslide and rockfall event occurs at the tunnel entrance, the video stream of the landslide and rockfall detection at the tunnel entrance continues to be read.

[0013] Step 9, the S1022 intelligent early warning algorithm for geological disasters caused by flooding of railways, which is part of the S102 intelligent algorithm training function mentioned in Step 5, is characterized by analyzing the collected video data of flooded railways and generating an early warning strategy based on the analysis results, including the following steps: Using cameras to capture video streams of railway flooding detection; The video data was analyzed using a GIS (Geographic Information System)-based method for calculating flood inundation in complex terrain and a flood-inundated railway event detection algorithm based on continuous comparison of high-definition video images. The analysis results were used to determine whether a flood-inundated railway event occurred. If a railway flooding event occurs, an early warning strategy is generated based on the type of the event and the location where it is detected. If no railway flooding event occurs, the system continues to read the railway flooding detection video stream.

[0014] Step 10, the S1023 overhead power line foreign object intelligent early warning algorithm in the S102 intelligent algorithm training function mentioned in Step 5, is characterized by analyzing the collected video data of foreign objects on the overhead power line and generating an early warning strategy based on the analysis results, including the following steps: Using cameras to capture video streams of foreign object detection on overhead power lines; A deep learning-based foreign object event detection algorithm for overhead power lines is used to analyze video data, and the analysis results are used to determine whether there are foreign object events on overhead power lines. If an overhead power line foreign object event is detected, an early warning strategy is generated based on the type of the event and its location. If no such event is detected, the overhead power line foreign object detection video stream continues to be read.

[0015] Step 11, the S1024 intelligent algorithm for personnel or large animal track encroachment in the S102 intelligent algorithm training function described in Step 5, is characterized by analyzing the collected video data of personnel or large animal track encroachment and generating an early warning strategy based on the analysis results, including the following steps: Using cameras to collect video streams for detecting encroachment on tracks by people or large animals; The algorithm for detecting human or large animal encroachment on the track is based on machine vision deep learning to analyze video data and determine whether there are any such events based on the analysis results. If a human or large animal encroachment incident occurs, an early warning strategy is generated based on the type of the incident and its detected location; if no human or large animal encroachment incident occurs, the video stream of the detected human or large animal encroachment incident continues to be read.

[0016] Step 12, the S1025 intelligent smoke and fire alarm early warning algorithm in the S102 intelligent algorithm training function mentioned in Step 5, is characterized by analyzing the collected smoke and fire alarm video data and generating an early warning strategy based on the analysis results, including the following steps: Use cameras to capture video streams of smoke and fire alarm detection; The algorithm for detecting smoke and fire alarms uses deep learning based on machine vision technology, combined with smoke sensors, to analyze video data and determine whether a smoke and fire alarm has occurred based on the analysis results. If a fire alarm event is detected, an early warning strategy is generated based on the type of fire alarm event and the location where the fire alarm event is detected; if no fire alarm event is detected, the fire alarm detection video stream is read again.

[0017] Step 13, the S1026 intelligent early warning algorithm for platform personnel encroachment in the S102 intelligent algorithm training function mentioned in Step 5, is characterized by analyzing the collected video data of platform personnel encroaching on platform boundaries and generating an early warning strategy based on the analysis results, including the following steps: Using cameras to collect video streams of people encroaching on platform limits while waiting for trains; The video data was analyzed using a deep learning-based algorithm to detect instances of passengers encroaching on platform boundaries, both inside and outside the yellow lines. Based on the analysis results, it was determined whether such instances occurred. If a platform passenger encroachment incident is detected, an early warning strategy is generated based on the type of the incident and its location. If no such incident is detected, the system continues to read the platform passenger encroachment detection video stream.

[0018] Step 14, the S1027 intelligent early warning algorithm for platform personnel falls in the S102 intelligent algorithm training function described in Step 5, is characterized by analyzing the collected video data of platform personnel falls and generating an early warning strategy based on the analysis results, including the following steps: Using cameras to collect video streams of people falling on the platform; The algorithm for detecting platform personnel falls is based on deep learning and machine vision. Video data is analyzed, and the results are used to determine whether there are any platform personnel fall incidents. If a platform personnel fall incident is detected, an early warning strategy is generated based on the type of platform personnel fall incident and the location where the incident was detected; if no platform personnel fall incident is detected, the platform personnel fall detection video stream is read again.

[0019] Step 15, the S1028 debris flow intelligent early warning algorithm in the S102 intelligent algorithm training function mentioned in Step 5, is characterized by analyzing the collected debris flow video data and generating an early warning strategy based on the analysis results, including the following steps: Using cameras to capture video streams of debris flow detection; A debris flow event detection algorithm based on deep learning methods for calculating debris flow flow rate and velocity is used to analyze video data, and the existence of debris flow events is determined based on the analysis results. If a debris flow event is detected, an early warning strategy is generated based on the debris flow event type and the location where the debris flow event is detected; if no debris flow event is detected, the debris flow detection video stream is read again.

[0020] Step 16, the S1029 intelligent early warning algorithm for monitoring crowd congestion at the exit, which is part of the S102 intelligent algorithm training function described in Step 5, is characterized by analyzing the collected video data of crowd congestion at the exit and generating an early warning strategy based on the analysis results, including the following steps: Use cameras to collect video streams of crowd control at the station exit; An algorithm for detecting crowding events at station exits is used to analyze video data based on the theory of passenger flow queuing in video images to measure the congestion level in a specified area, and the analysis results are used to determine whether there are crowding events at the station exits. If a crowding event occurs at the exit, an early warning strategy is generated based on the type of crowding event and the location where the crowding event is detected; if no crowding event occurs at the exit, the video stream of the crowding detection at the exit continues to be read.

[0021] The types of intelligent algorithm training functions described in step 5 are not limited to those listed in steps 7-16 above. All artificial intelligence algorithms that are beneficial to the safety of railway line patrol can be applied by loading them onto the edge computer in the railway line intelligent patrol system cloud platform S1. Detailed Implementation The present invention will now be described in detail with reference to the accompanying drawings. The specific operating methods in the embodiments can also be applied to the device embodiments or system embodiments.

[0022] Step 1: This invention provides an intelligent early warning system for railway line patrol safety based on edge computing. The system includes: a railway line intelligent patrol system cloud platform S1, a railway line intelligent patrol system network infrastructure unit S2, an on-site edge intelligent early warning device and traffic control unit S3, and an Internet of Things (IoT) sensing facility unit S4; (Reference) Figure 1 ; The intelligent railway patrol system cloud platform S1 includes the China Railway Administration central cloud platform S10 and regional bureau cloud platforms S11. The central cloud platform S10 manages all regional cloud platforms, monitors the safety status of railway lines across the entire railway network, analyzes the impact of abnormal events on the overall railway network operation, and makes timely decisions to allocate resources, minimizing the negative impacts of various anomalies. The regional bureau cloud platform S11 manages all intelligent early warning devices within its region, monitors railway line patrols within its region, analyzes the causes and countermeasures of abnormal events within its region, and takes necessary emergency measures. (Reference) Figure 4 , Figure 5 ; The intelligent patrol system network infrastructure unit S2 along the railway line includes the China Railway Communication private network and the public networks of the three major telecom operators along the railway line, as well as the station local area networks built on these two major network communication facilities, and their power supply guarantee facilities S20, etc., and all facilities and methods to ensure network security and stable operation; (Reference) Figure 3 ; The on-site edge intelligent early warning device and traffic control unit S3 includes all edge computers installed at key monitoring locations, alarm systems requiring real-time control, other automated emergency equipment, and power supply protection equipment for the aforementioned facilities; this unit is the core facility for achieving the purpose of this invention and a key equipment unit for ensuring that various abnormal events are handled correctly and effectively in real time; Reference Figure 2 ; The IoT sensing facility unit S4 includes all IoT video sensing facility units along the railway line, as well as other sensing equipment required for line patrol, such as radio frequency sensors for precise control of carriage position sensing; (Reference) Figure 2 .

[0023] Step 2: The network video camera S41 used in the IoT video sensing facility unit S4 mentioned in Step 1 is a Hikvision DS-2CD3T87WDA3-LS 8-megapixel camera in this example. The power supply and backup power system S42 ensure its normal operation. Since railway lines often traverse remote areas with complex geological and climatic conditions, the power supply unit must consider backup photovoltaic and battery power systems to maintain power supply for a period of time in case of problems with the normal power supply facilities. In this example, S42 adopts the wind-solar hybrid power supply system solution from Guangdong Weilan New Energy, consisting of a power generation unit, a control unit, and an energy storage unit, ensuring that the video camera can upload a video stream of a certain duration for event analysis even when wired power is lost. (Reference) Figure 2 .

[0024] Step 3: The field edge computer S31 of the intelligent early warning device and traffic control unit S3 mentioned in Step 1 uses the B400 series model from Hefei Lianbao Technology Co., Ltd. The outdoor power management unit S30, like the video camera in Step 2, consumes more power due to its similar environmental characteristics. Therefore, a more powerful power management unit is needed. In this example, S30 uses the wind-solar hybrid power supply system solution from Guangdong Weilan New Energy, consisting of a power generation unit, a control unit, and an energy storage unit. This ensures the edge computer can maintain a certain operating time, analyze and judge abnormal events in various video streams, and accurately and promptly upload data to the regional cloud platform. (Reference) Figure 2 ; The Lianbao B400 edge computer S31 is equipped with an edge computing platform independently developed by Shenzhen Yuncun Technology Co., Ltd., and features functions such as on-site IPC video device management (S311), task orchestration management (S312), algorithm repository management (S313), network settings management (S315), system settings management (S316), alarm upload and on-site alarm management, and anomaly policy management. (Reference) Figure 2 ; The S311 field IPC video device management function features: a field edge computer enables real-time video browsing and camera management for viewing various devices on-site; a video list manages video browsing, viewing, editing, and deletion from cameras; and it can also add and manage other IPC video devices; (See reference) Figure 2 ; The task orchestration and management function S312 includes task orchestration and task recording. Devices are orchestrated and managed in the task center. When adding a task, steps such as "task name," "recognition interval," "camera selection," "algorithm loading," and "filtering mode" are completed. After selecting the camera, the required algorithm is loaded, and the filtering mode is selected. During configuration, algorithms can be freely plugged in and out. The bounding box of the target object can be selected as needed: "completely within the ROI," "center point within the ROI," or "overlapping with the ROI." Other values ​​will be filtered and not displayed in the task record. The task record records and traces the events in which the device algorithm takes effect; the record list corresponds to the task list. (Reference) Figure 2 , Figure 5 ; The algorithm repository management function S313 allows users to view authorized algorithms in the system's algorithm repository. Authorized algorithms are distributed from the regional cloud platform to the edge computing platform for authorization, followed by task management and other operations. (See reference...) Figure 5 ; The upload alarm and on-site alarm management function S314 triggers the edge computer platform to report abnormal events to the higher-level regional cloud platform if an abnormal event occurs in the video stream for various functions and purposes. Simultaneously, it uploads relevant video evidence, triggers on-site alarms to issue alarm signals, and initiates necessary temporary emergency measures. The on-site emergency event intelligent control and processing function S318 minimizes the impact of abnormal events as much as possible. (See reference...) Figure 2 , Figure 5 ; The S315 network settings management tool allows you to configure or modify network settings, default gateway, IP address, subnet mask, and DNS. After making changes, click "OK" to save the configuration. (See reference...) Figure 2 ; The S316 system configuration function displays the device name, allows modification, and saves the configuration. The device time has already been synchronized, either via NTP (Network Time Protocol) or manually, before saving the configuration. (See reference...) Figure 2 ; The exception policy management function S317 includes rule management and rule recording. Rule management manages added tasks, including ID, rule name, rule template, rule status, and creation time, allowing for viewing and deletion. Adding a rule includes information such as "rule name," "task source," "rule template," and "remarks." The task source is selected based on the task name created in the task center. The rule template can be selected based on target dwell timeout, target quantity timeout threshold, or target long-term loss. After adding remarks, clicking "Next" allows selecting the ROI. The ROI selection box is based on the ROI number created in the task center. After selecting the event name, click Next to set the ROI. The ROI settings are based on the rule template, which considers both time and quantity. The target dwell timeout and target long-term loss options allow for customizable timeout durations (in seconds), and the target quantity timeout threshold allows for customizable identification counts (in units). The rule record records events that have taken effect under rule management. The rule record includes ID, rule name, task name, camera name, rule template, occurrence time, and thumbnail. Rule records can be viewed and reviewed by querying the rule name, task name, camera name, and occurrence time. (Reference) Figure 2 .

[0025] Step 4: The network facility unit S2 of the intelligent patrol system along the railway line mentioned in Step 1 has an underlying infrastructure consisting of a local area network (LAN) composed of Hefei Lianbao Technology Co., Ltd. B400 edge computers, several Hikvision video cameras, and network devices (routers, industrial gateways, etc.) at key locations along the railway line. If the network devices are outdoors, they also adopt the wind-solar hybrid power supply system solution of Guangdong Weilan New Energy, which includes a power generation unit, a control unit, and an energy storage unit, providing a powerful power management unit (on-site network device power management) S20. The power supply mode for these network devices indoors can be a backup power supply such as a UPS power supply to ensure that the network infrastructure such as routers and gateways can maintain a certain working time and ensure that the edge computers can reliably transmit information to the cloud platform when abnormal events occur. The LAN can be composed of a combination of wired network and wireless WiFi network to meet the on-site networking needs of various heterogeneous devices. Several m LANs based on edge computers gather information to the regional bureau cloud platform through the railway communication network. The railway communication network includes China Railcom's proprietary network system S21 and the public network system S22 of the three major operators (China Mobile, China Unicom, China Telecom, etc.). China Railcom's proprietary network system S21 includes the railway communication 4G / 5G wireless network S23 and the railway communication dedicated wired network S24. The wired network is a dedicated network for normal use. When the wired network fails due to natural disasters, damage to the main power supply line, or other reasons, the wireless network serves as a supplementary backup network, or the 4G / 5G wireless network can be used directly in places where it is inconvenient to lay cables. The public network system S22 of the three major operators (China Mobile, China Unicom, China Telecom, etc.) is a supplementary backup network for this railway dedicated line network, and also includes the 4G / 5G wireless network S263 and the wired network along the railway line S25. The infrastructure of the railway center network in the intelligent patrol system network facility unit S2 along the railway line is also based on the principle of using China Tietong's dedicated network as the main network and the networks of the three major telecom operators as supplementary networks; (Refer to...) Figure 3 .

[0026] Step 5: The S1 infrastructure of the intelligent patrol system cloud platform along the railway line mentioned in Step 1, on the one hand, establishes a dedicated cloud server of China Railcom to maintain the normal operation of the cloud platform, and on the other hand, other commercial servers can be used as backup means to ensure the reliable operation of the information link. The General Administration's central cloud platform S10 includes central control functions S101, intelligent algorithm training functions S102, security module functions S103, cloud computing functions S104, and intelligent analysis and decision-making functions S105; it monitors the status of all railway lines in the national railway system, collects information from n regional bureau cloud platforms, and references... Figure 4 ; The central control function S101 includes network control, security control, computing power control, data control, node control, and control of various scenarios, etc., automatically monitoring the status of railway lines and automatically optimizing control measures; The intelligent algorithm training function S102 is a reliable source of algorithms for various scenarios. Only the China Railway Administration possesses comprehensive direct data on various scenarios, boasting unparalleled big data conditions that no other company or social organization can match. This forms a solid foundation for optimizing and training models for various scenario algorithms. Currently, the urgently needed scenario algorithms are as follows: Intelligent early warning algorithm for crossing railings S1020; Intelligent early warning algorithm for geological disasters such as landslides and rockfalls at tunnel entrances S1021; Intelligent early warning algorithm for geological disasters such as flooding of railways S1022; Intelligent early warning algorithm for foreign objects on overhead power lines S1023; Intelligent algorithm for personnel or large animals encroaching on rail limits S1024; Intelligent early warning algorithm for smoke and fire alarms S1025; Intelligent early warning algorithm for platform personnel encroaching on rail limits S1026; Intelligent early warning algorithm for platform personnel falling S1027; Intelligent early warning algorithm for debris flow S1028; Intelligent early warning algorithm for monitoring crowd congestion at station exits S1029, etc. Security module S103 refers to firewalls, antivirus, intrusion detection, intrusion prevention, log auditing, network gateways, etc. As the main artery of international economic operation, the security of the railway platform is extremely important. Any input and output of information data must be isolated and processed by security modules to meet national security level requirements. The cloud computing function S104 is the computing power within the cloud platform designed to meet the needs of data analysis. The intelligent analysis and decision-making function S105 is a manifestation of the role of the cloud platform. All the data acquired is for the cloud platform, which acts as the brain, to perform rapid calculations on the server and provide correct decision-making references based on the correct rules and constraints of railway operation. These references are then pushed to the railway dispatching system S12 for train operation scheduling reference. The regional bureau cloud platform S11 includes a regional cloud algorithm warehouse function S111, a powerful regional center control function S112, a cloud-edge high-efficiency collaboration function S113, and a security module S114; it monitors the status of all edge computers within the regional bureau, obtains the status of railway lines within the region from the edge computers, and actively pushes it to the central cloud platform of the headquarters; (Reference) Figure 5 ; The algorithms in the regional cloud algorithm warehouse function S111 are mostly provided by the intelligent algorithm training function center S102 of the China Railway Corporation cloud platform, and a small part of the algorithms can come from artificial intelligence algorithm companies other than the China Academy of Railway Sciences. The powerful regional central control S112 optimizes the functional application distribution of edge computers within the region, rationally manages computing resources, arranges various algorithm function requirements in places where they are urgently needed, manages network node data flow, and ensures reasonable allocation of network resources. The cloud-edge high-efficiency collaboration function S113 refers to the integration of computing power, storage, network and other resources of edge computers into servers, helping users to quickly build edge computing cloud network environments, supporting the network and computing power needs of intelligent applications at low cost and high efficiency; regional cloud realizes unified orchestration, scheduling, configuration and management of complex network, computing power, data, application and other resources on the edge side, shielding the complexity of the edge side, and providing the edge side with AI, IoT, security and other capabilities support; Security module S114, like S103 of the central cloud platform, ensures that only edge computers authorized by the regional bureau platform can access the cloud platform, and isolates unauthorized devices from intruding into the regional cloud platform system.

[0027] Step 6: Most of the scene algorithms in the intelligent algorithm training function S102 described in Step 5 are artificial intelligence algorithms based on video streams. Their basic workflow diagrams are roughly the same; the process is roughly as follows: P000 The central cloud platform of the intelligent patrol system along the railway line provides various intelligent early warning algorithms approved by the General Administration; P100 The cloud platform of the railway regional bureau receives the intelligent early warning algorithms authorized by the General Administration platform; P110 The cloud platform of the railway regional bureau distributes various intelligent early warning algorithms to the edge computer units with different algorithm requirements within the region; P120 The edge computer units assign the intelligent early warning algorithms to various video stream sources with different application requirements. The video stream sources come from cameras installed at different application requirement locations; P130 Various video streams that meet the requirements of different algorithms are obtained from various cameras; P140 Does the intelligent algorithm identify the assigned abnormal situation from the associated video stream? If no abnormal event is identified, the process returns to P130 to continue working. If an abnormal event that meets the algorithm's trigger requirements is identified, P150 records the abnormal event on the local edge computer and initiates on-site emergency measures. P160 intelligently processes on-site warnings and alerts, reports the abnormal event to the cloud platform, and pushes relevant video streams. P170 continues to acquire various video streams from different cameras that meet different algorithm requirements. P180 checks if the intelligent algorithm has identified the assigned abnormal situation from the associated video streams. If it is still an abnormal situation, the process returns to P170 to continue working. If the abnormal event has been identified and processed, P190 deactivates the on-site alarm settings, allowing manual restoration of emergency measures to normal status. P200 sends a message to the cloud platform center that the abnormal event has been processed and the alarm has been deactivated, then returns to P130 to continue monitoring the railway line status. (Reference) Figure 6 .

[0028] Step 7, the intelligent algorithm training function S1020 for hurdle crossing intelligent early warning algorithm described in step 5, is characterized by analyzing the collected hurdle crossing video data and generating an early warning strategy based on the analysis results, including the following steps: Use a camera to capture video streams of objects crossing railings; A deep learning-based fence crossing event detection algorithm is used to analyze video data, and the existence of fence crossing events is determined based on the analysis results. If a fence crossing event is detected, an early warning strategy is generated based on the fence crossing event type and the location where the fence crossing event was detected; if no fence crossing event is detected, the fence crossing detection video stream is read again.

[0029] Step 8, the intelligent algorithm training function S1021 for geological disasters such as landslides and rockfalls at tunnel entrances, described in Step 5, is characterized by analyzing the collected video data of landslides and rockfalls at tunnel entrances and generating early warning strategies based on the analysis results, including the following steps: Using cameras to capture video streams of landslide and rockfall detection at tunnel entrances; A tunnel entrance landslide and rockfall event detection algorithm based on kinematics and the law of conservation of energy is used to analyze video data, and the existence of a tunnel entrance landslide and rockfall event is determined based on the analysis results. If a landslide and rockfall event occurs at the tunnel entrance, an early warning strategy is generated based on the type of landslide and rockfall event and the location where the event is detected. If no landslide and rockfall event occurs at the tunnel entrance, the video stream of the landslide and rockfall detection at the tunnel entrance continues to be read.

[0030] Step 9, the intelligent algorithm training function S1022 mentioned in Step 5, the intelligent early warning algorithm for geological disasters caused by flooding of railways, is characterized by analyzing the collected video data of flooded railways and generating early warning strategies based on the analysis results, including the following steps: Using cameras to capture video streams of railway flooding detection; The video data was analyzed using a GIS (Geographic Information System)-based method for calculating flood inundation in complex terrain and a flood-inundated railway event detection algorithm based on continuous comparison of high-definition video images. The analysis results were used to determine whether a flood-inundated railway event occurred. If a railway flooding event occurs, an early warning strategy is generated based on the type of the event and the location where it is detected. If no railway flooding event occurs, the system continues to read the railway flooding detection video stream.

[0031] Step 10, the intelligent algorithm training function S1023 overhead power line foreign object intelligent early warning algorithm mentioned in step 5, is characterized by analyzing the collected video data of foreign objects on the overhead power line and generating an early warning strategy based on the analysis results, including the following steps: Using cameras to capture video streams of foreign object detection on overhead power lines; A deep learning-based foreign object event detection algorithm for overhead power lines is used to analyze video data, and the analysis results are used to determine whether there are foreign object events on overhead power lines. If an overhead power line foreign object event is detected, an early warning strategy is generated based on the type of the event and its location. If no such event is detected, the overhead power line foreign object detection video stream continues to be read.

[0032] Step 11, the intelligent algorithm training function S1024 described in Step 5, is characterized by analyzing the collected video data of personnel or large animals encroaching on the track, and generating an early warning strategy based on the analysis results, including the following steps: Using cameras to collect video streams for detecting encroachment on tracks by people or large animals; The algorithm for detecting human or large animal encroachment on the track is based on machine vision deep learning to analyze video data and determine whether there are any such events based on the analysis results. If a human or large animal encroachment incident occurs, an early warning strategy is generated based on the type of the incident and its detected location; if no human or large animal encroachment incident occurs, the video stream of the detected human or large animal encroachment incident continues to be read.

[0033] Step 12, the intelligent algorithm training function S1025 for intelligent smoke and fire alarms described in Step 5, is characterized by analyzing the collected smoke and fire alarm video data and generating an early warning strategy based on the analysis results, including the following steps: Use cameras to capture video streams of smoke and fire alarm detection; The algorithm for detecting smoke and fire alarms uses deep learning based on machine vision technology, combined with smoke sensors, to analyze video data and determine whether a smoke and fire alarm has occurred based on the analysis results. If a fire alarm event is detected, an early warning strategy is generated based on the type of fire alarm event and the location where the fire alarm event is detected; if no fire alarm event is detected, the fire alarm detection video stream is read again.

[0034] Step 13, the intelligent algorithm training function S1026 platform passenger encroachment intelligent early warning algorithm mentioned in Step 5, is characterized by analyzing the collected platform passenger encroachment video data and generating an early warning strategy based on the analysis results, including the following steps: Using cameras to collect video streams of people encroaching on platform limits while waiting for trains; The video data was analyzed using a deep learning-based algorithm to detect instances of passengers encroaching on platform boundaries, both inside and outside the yellow lines. Based on the analysis results, it was determined whether such instances occurred. If a platform passenger encroachment incident is detected, an early warning strategy is generated based on the type of the incident and its location. If no such incident is detected, the system continues to read the platform passenger encroachment detection video stream.

[0035] Step 14, the intelligent algorithm training function S1027 platform personnel fall intelligent early warning algorithm mentioned in step 5, is characterized by analyzing the collected platform personnel fall video data and generating an early warning strategy based on the analysis results, including the following steps: Using cameras to collect video streams of people falling on the platform; The algorithm for detecting platform personnel falls is based on deep learning and machine vision. Video data is analyzed, and the results are used to determine whether there are any platform personnel fall incidents. If a platform personnel fall incident is detected, an early warning strategy is generated based on the type of platform personnel fall incident and the location where the incident was detected; if no platform personnel fall incident is detected, the platform personnel fall detection video stream is read again.

[0036] Step 15, the intelligent algorithm training function S1028 debris flow intelligent early warning algorithm mentioned in step 5, is characterized by analyzing the collected debris flow video data and generating an early warning strategy based on the analysis results, including the following steps: Using cameras to capture video streams of debris flow detection; A debris flow event detection algorithm based on deep learning methods for calculating debris flow flow rate and velocity is used to analyze video data, and the existence of debris flow events is determined based on the analysis results. If a debris flow event is detected, an early warning strategy is generated based on the debris flow event type and the location where the debris flow event is detected; if no debris flow event is detected, the debris flow detection video stream is read again.

[0037] Step 16, the intelligent algorithm training function S1029 described in Step 5, which is an intelligent early warning algorithm for monitoring crowd congestion at the exit, is characterized by analyzing the collected video data of crowd congestion at the exit and generating an early warning strategy based on the analysis results, including the following steps: Use cameras to collect video streams of crowd control at the station exit; An algorithm for detecting crowding events at station exits is used to analyze video data based on the theory of passenger flow queuing in video images to measure the congestion level in a specified area, and the analysis results are used to determine whether there are crowding events at the station exits. If a crowding event occurs at the exit, an early warning strategy is generated based on the type of crowding event and the location where the crowding event is detected; if no crowding event occurs at the exit, the video stream of the crowding detection at the exit continues to be read.

[0038] The types of intelligent algorithm training functions described in step 5 are not limited to those listed in steps 7-16 above. All artificial intelligence algorithms that are beneficial to the safety of railway line patrol can be applied by loading them onto the edge computer in the railway line intelligent patrol system cloud platform S1.

[0039] The beneficial effects of this invention are: This invention deploys edge computing technology along railway lines, enabling data processing and analysis directly near the data source, reducing data transmission latency, achieving real-time monitoring of railway conditions and rapid identification of anomalies, effectively shortening the time from detection to response, and significantly improving the efficiency and accuracy of railway patrol safety emergency response; This invention utilizes distributed edge computing capabilities to provide real-time feedback of railway information to the regional cloud platform. Trains can then promptly access railway information along the line from the cloud platform, gaining accurate insights into railway conditions in advance, thus preventing major safety accidents and improving railway operational efficiency and capacity.

[0040] This invention achieves precise monitoring and early warning of key areas by deploying a distributed edge intelligent early warning subsystem at critical railway locations. This targeted deployment strategy not only reduces the construction and operation costs of the system but also improves train safety. Attached Figure Description Figure 1 This is a framework diagram of a railway line patrol safety intelligent early warning system based on edge computing; Figure 2 This diagram shows the composition of intelligent early warning devices, traffic control units, and IoT video sensing facilities at key locations along the railway line. Figure 3 This is a diagram showing the network infrastructure units of the intelligent patrol system along the railway line; Figure 4 This is a diagram showing the components of the General Administration's central cloud platform for the intelligent patrol system along railway lines. Figure 5 This is a cloud platform and video data flow diagram of the intelligent patrol system along the railway line. Figure 6 This is a basic flowchart of the operation of various algorithms in the intelligent patrol system along the railway line.

Claims

1. Claim 1: The intelligent early warning system for railway line patrol safety based on edge computing comprises: The intelligent patrol system along the railway line consists of a cloud platform S1, a network facility unit S2, an on-site edge intelligent early warning device and traffic control unit S3, and an Internet of Things sensing facility unit S4; see Figure 1 for reference. The intelligent patrol system cloud platform S1 along the railway line includes the China Railway Administration central cloud platform S10 and regional bureau cloud platforms S11. The central cloud platform S10 manages all regional cloud platforms, controls the safety status of railway lines across the entire railway network, analyzes the impact of abnormal events on the overall railway network operation, and makes timely decisions to allocate resources, minimizing the negative impacts of various anomalies. The regional bureau cloud platform S11 manages all intelligent early warning devices within its region, monitors railway line patrols within its region, analyzes the causes and countermeasures of abnormal events within its region, and takes necessary emergency measures. (Refer to Figures 4 and 5.) The intelligent patrol system network facility unit S2 along the railway line includes the China Railway Communication private network and the public networks of the three major operators along the railway line, as well as the station local area network built on these two major network communication facilities, and its power supply guarantee facilities S20, etc., and all facilities and methods to ensure network security and stable operation; see Figure 3; The on-site edge intelligent early warning device and traffic control unit S3 include all edge computers installed at key monitoring locations, alarm systems requiring real-time control, other automated emergency equipment, and power supply protection equipment for the above facilities; this unit is the core facility for achieving the purpose of this invention and a key equipment unit to ensure that various abnormal events are handled correctly and effectively in real time; see Figure 2; The IoT sensing facility unit S4 includes all IoT video sensing facility units along the railway line, as well as other sensing devices required for line patrol, such as radio frequency sensors for precise control of carriage position perception; see Figure 2.

2. Claim 2: The IoT video sensing facility unit S4 described in claim 1 is characterized by including a network video camera S41 and a power supply and backup power supply system S42 to ensure its normal operation. Since railway lines cross mountains, rivers, and other remote areas with complex geological and climatic conditions, the power supply unit must consider the backup photovoltaic and battery power supply systems to maintain power supply for a period of time in the event of a problem with the normal power supply facilities, so as to ensure that the video camera can upload a video stream of a certain length for event analysis in a timely manner; see Figure 2.

3. Claim 3: The field edge intelligent device power management unit S30 of the field edge intelligent early warning device and traffic control unit S3 described in claim 1 has the same environmental characteristics as the video camera in claim 2, and consumes more power. Therefore, a more powerful power management unit is needed to ensure that the edge computer can maintain a certain working time, analyze and judge the abnormal events in various video streams, and accurately and timely upload them to the regional cloud platform; see Figure 2; The field edge computer S31 must have the following functions: field IPC video device management S311, task orchestration management S312, algorithm repository management S313, network settings management S315, system settings management S316, alarm upload and field alarm management, and anomaly policy management; see Figure 2. The on-site IPC video device management function S311 is characterized by being on-site The edge computer enables real-time video browsing and camera management for viewing various devices on-site. The video list manages video browsing, viewing, editing, and deletion from cameras, and can also add and manage other IPC video devices; see Figure 2. The task orchestration management function S312 includes task orchestration and task recording. Devices are orchestrated and managed in the task center. When adding a task, steps such as "task name," "recognition interval," "camera selection," "algorithm loading," and "filtering mode" are completed. After selecting the camera, the required algorithm is loaded, and the filtering mode is selected. During the configuration process, algorithms can be freely plugged in and called. The bounding box of the target object can be selected as "completely within the ROI," "center point within the ROI," or "overlapping with the ROI." Other options will be filtered and not displayed in the task record. The task record records and traces back the device algorithm activation events, and the record list corresponds to the task list; see Figures 2 and 5. The algorithm repository management function S313 allows users to view authorized algorithms in the system's algorithm repository. Authorized algorithms are distributed from the regional cloud platform to the edge computer platform for authorization, followed by task management and other operations; see Figure 5. The aforementioned alarm upload and on-site alarm management function S314 is designed to trigger the edge computer platform to report abnormal events to the higher-level regional cloud platform if abnormal events occur in video streams with different functions and purposes. At the same time, it uploads relevant video evidence, triggers the on-site alarms to issue alarm signals, and triggers the necessary temporary emergency measures to be linked up. The on-site emergency event intelligent control and processing function S318 minimizes the impact of abnormal events as much as possible; see Figures 2 and 5. The network settings management S315 is used to set or modify network settings, default gateway, IP address, subnet mask, and DNS. After making changes, click OK to save the configuration; see Figure 2. The system configuration function S316 displays the device name, modifies it, and saves the configuration. The device time has been synchronized, either via NTP (Network Time Protocol) or manually, and then the configuration is saved; see Figure 2. The anomaly strategy management function S317 is rule management and rule recording. Rule management manages added tasks, including ID, rule name, rule template, rule status, and creation time, allowing for viewing and deletion. Adding a rule includes information such as "rule name," "task source," "rule template," and "remarks." The task source is selected based on the task name created in the task center. The rule template can be selected based on target dwell timeout, target quantity timeout threshold, or target long-term loss. After adding remarks, clicking "Next" allows selecting the ROI. The ROI selection box is based on the ROI number created in the task center. After selecting the event name, click Next to set the ROI. The ROI settings are based on the rule template, which considers both time and quantity. The target dwell timeout and target long-term loss options can be customized in seconds, and the target quantity timeout threshold can be customized in terms of the number of targets to be identified. The rule record records the effective events of the rule management. The rule record includes ID, rule name, task name, camera name, rule template, occurrence time, and thumbnail. You can view and trace the rule record by querying the rule name, task name, camera name, and occurrence time; see Figure 2.

4. Claim 4: The network facility unit S2 of the intelligent patrol system along the railway line described in claim 1 is not an independent implementation. Its underlying facility is a local area network composed of edge computers at key locations along the railway line, several video cameras (or other railway signal sensing sensors), and network devices (routers, industrial gateways, etc.). These facilities also need to provide a powerful power management unit (field network device power management) S20, similar to the environmental characteristics of the video cameras in claim 2, to ensure that the network infrastructure such as routers and gateways maintains a certain working time and that the edge computers can reliably transmit information to the cloud platform when abnormal events occur. The local area network can be composed of a combination of wired network and wireless WiFi network to meet the field networking needs of various heterogeneous devices. Several m local area networks based on edge computers gather information to the regional bureau cloud platform through the railway communication network. The railway communication network includes China Railcom's proprietary network system S21 and the public network system S22 of the three major operators (China Mobile, China Unicom, China Telecom, etc.). The China Railcom proprietary network system S21 includes a railway communication 4G / 5G wireless network S23 and a railway communication dedicated wired network S24. The wired network is a dedicated network used normally. When the wired network fails due to natural disasters, damage to the main power supply line, or other reasons, the wireless network serves as a supplementary backup network, or the 4G / 5G wireless network can be used directly in places where it is inconvenient to lay cables. The public network system S22 of the three major operators (China Mobile, China Unicom, China Telecom, etc.) is a supplementary backup network for this railway dedicated line network, and also includes the 4G / 5G wireless network S263 and the wired network along the railway line S25. The infrastructure of the railway center network of the intelligent patrol system network facility unit S2 along the railway line is also based on the principle of using China Tietong's dedicated network as the main network and the networks of the three major operators as supplementary networks; see Figure 3.

5. Claim 5: The cloud platform S1 infrastructure of the intelligent patrol system along the railway line described in claim 1, on the one hand, establishes a dedicated cloud server of China Railcom to maintain the normal operation of the cloud platform, and on the other hand, can also use other commercial servers as backup means to ensure the reliable operation of the information link; The General Administration's central cloud platform S10 includes a central control function S101, an intelligent algorithm training function S102, a security module function S103, a cloud computing function S104, and an intelligent analysis and decision-making function S105; it can control the status of all railway lines in the national railway system and monitor and collect information from n regional bureau cloud platforms, as shown in Figure 4. The central control function S101 includes network control, security control, computing power control, data control, node control, and control of various scenarios, etc., automatically monitoring the status of railway lines and automatically optimizing control measures; The intelligent algorithm training function S102 is a reliable source of algorithms for various scenarios. Only the China Railway Administration possesses comprehensive direct data on various scenarios, boasting unparalleled big data conditions that no other company or social organization can match. This forms a solid foundation for optimizing and training models for various scenario algorithms. Currently, the urgently needed scenario algorithms are as follows: Intelligent early warning algorithm for crossing railings S1020; Intelligent early warning algorithm for geological disasters such as landslides and rockfalls at tunnel entrances S1021; Intelligent early warning algorithm for geological disasters such as flooding of railways S1022; Intelligent early warning algorithm for foreign objects on overhead power lines S1023; Intelligent algorithm for personnel or large animals encroaching on rail limits S1024; Intelligent early warning algorithm for smoke and fire alarms S1025; Intelligent early warning algorithm for platform personnel encroaching on rail limits S1026; Intelligent early warning algorithm for platform personnel falling S1027; Intelligent early warning algorithm for debris flow S1028; Intelligent early warning algorithm for monitoring crowd congestion at station exits S1029, etc. The security module S103 refers to firewalls, antivirus, intrusion detection, intrusion prevention, log auditing, network gateways, etc. As the main artery of international economic operation, the security of the railway platform is extremely important. Any input and output of information data must be isolated and processed by the security module to meet national security level requirements. The cloud computing function S104 is the computing power within the cloud platform to meet the needs of data analysis. The intelligent analysis and decision-making function S105 is a manifestation of the role of the cloud platform. All the data acquired is for the cloud platform, which acts as the brain, to undergo rapid calculation by the server and provide correct decision-making references based on the correct rules and constraints of railway operation. These references are then pushed to the railway dispatching system S12 for train operation scheduling reference. The regional bureau cloud platform S11 includes a regional cloud algorithm warehouse function S111, a powerful regional center control function S112, a cloud-edge high-efficiency collaboration function S113, and a security module S114; it monitors the status of all edge computers within the regional bureau, obtains the status of railway lines within the region from the edge computers, and actively pushes it to the central cloud platform of the headquarters; see Figure 5. Most of the algorithms in the regional cloud algorithm warehouse function S111 are provided by the intelligent algorithm training function center S102 of the cloud platform of the China Railway Corporation, and a small part of the algorithms may come from artificial intelligence algorithm companies other than the China Academy of Railway Sciences. The powerful regional central control S112 optimizes the functional application distribution of edge computers within the region, rationally manages computing resources, arranges various algorithm function requirements in places where they are urgently needed, manages network node data flow, and ensures reasonable allocation of network resources. The cloud-edge high-efficiency collaboration function S113 refers to the integration of computing power, storage, network and other resources of edge computers into servers, helping users to quickly build an edge computing cloud network environment, supporting the network and computing power needs of intelligent applications at low cost and high efficiency; the regional cloud realizes unified orchestration, scheduling, configuration and management of complex network, computing power, data, application and other resources on the edge side, shielding the complexity of the edge side, and providing the edge side with AI, IoT, security and other capabilities support; The security module S114 is the same as S103 of the central cloud platform, ensuring that only edge computers authorized by the regional bureau platform can access the cloud platform, and isolating unauthorized devices from intruding into the regional cloud platform system.

6. Claim 6: The various scenario algorithms of the intelligent algorithm training function S102 described in claim 5 are mostly artificial intelligence algorithms based on video streams, and their basic workflow diagrams are roughly the same; the process is roughly as follows: P000 The central cloud platform of the intelligent patrol system along the railway line provides various intelligent early warning algorithms recognized by the General Administration; P100 The cloud platform of the railway regional bureau receives the intelligent early warning algorithms authorized by the General Administration platform; P110 The cloud platform of the railway regional bureau distributes various intelligent early warning algorithms to the edge computer units of different algorithm requirements within the region; P120 The edge computer unit assigns the intelligent early warning algorithms to various video stream sources with different application requirements, and the video stream sources come from cameras installed at different application requirement locations; P130 Various video streams that meet different algorithm requirements are obtained from various different cameras; P140 Does the intelligent algorithm identify the assigned abnormal situation from the associated video stream? If no abnormal event is identified, the process returns to P130 to continue working. If an abnormal event that meets the algorithm trigger criteria is identified, P150 records the abnormal event on the local edge computer and initiates on-site emergency measures. P160 intelligently processes on-site warnings and alerts, reports the abnormal event to the cloud platform, and pushes relevant video streams. P170 continues to acquire various video streams from different cameras that meet the requirements of different algorithms. P180 checks if the intelligent algorithm has identified the assigned abnormal situation from the associated video streams. If it is still an abnormal situation, the process returns to P170 to continue working. If the abnormal event has been identified and processed, P190 cancels the on-site alarm settings, allowing manual restoration of emergency measures to normal status. P200 sends a message to the cloud platform center that the abnormal event has been processed and the alarm has been canceled, then returns to P130 to continue monitoring the railway line status and acquires various video streams from different cameras that meet the requirements of different algorithms. (Refer to Figure 6.) 7. Claim 7: The S1020 intelligent early warning algorithm for crossing the hurdle in the S102 intelligent algorithm training function of claim 5, characterized in that... Analyzing the collected video data of people crossing the railing and generating early warning strategies based on the analysis results includes the following steps: Use a camera to capture video streams of objects crossing railings; A deep learning-based fence crossing event detection algorithm is used to analyze video data, and the existence of fence crossing events is determined based on the analysis results. If a fence crossing event is detected, an early warning strategy is generated based on the fence crossing event type and the location where the fence crossing event was detected; if no fence crossing event is detected, the fence crossing detection video stream is read again.

8. Claim 8: The S1021 intelligent early warning algorithm for geological disasters such as landslides and rockfalls at tunnel entrances in the S102 intelligent algorithm training function of claim 5, characterized in that... Analyzing the collected video data of landslides and rockfalls at the tunnel entrance, and generating early warning strategies based on the analysis results, includes the following steps: Using cameras to capture video streams of landslide and rockfall detection at tunnel entrances; A tunnel entrance landslide and rockfall event detection algorithm based on kinematics and the law of conservation of energy is used to analyze video data, and the existence of a tunnel entrance landslide and rockfall event is determined based on the analysis results. If a landslide and rockfall event occurs at the tunnel entrance, an early warning strategy is generated based on the type of landslide and rockfall event and the location where the event is detected. If no landslide and rockfall event occurs at the tunnel entrance, the video stream of the landslide and rockfall detection at the tunnel entrance continues to be read.

9. Claim 9: The S1022 intelligent early warning algorithm for geological disasters caused by flooding of railways, which is part of the S102 intelligent algorithm training function described in claim 5, is characterized by: Analyzing the collected video data of flooded railways and generating early warning strategies based on the analysis results includes the following steps: Using cameras to capture video streams of railway flooding detection; The video data was analyzed using a GIS (Geographic Information System)-based method for calculating flood inundation in complex terrain and a flood-inundated railway event detection algorithm based on continuous comparison of high-definition video images. The analysis results were used to determine whether a flood-inundated railway event occurred. If a railway flooding event occurs, an early warning strategy is generated based on the type of the event and the location where it is detected. If no railway flooding event occurs, the system continues to read the railway flooding detection video stream.

10. Claim 10: The intelligent early warning algorithm for foreign objects on overhead power lines in the S102 intelligent algorithm training function of claim 5, characterized in that... Analyzing the collected video data of foreign objects on overhead power lines and generating early warning strategies based on the analysis results includes the following steps: Using cameras to capture video streams of foreign object detection on overhead power lines; A deep learning-based foreign object event detection algorithm for overhead power lines is used to analyze video data, and the analysis results are used to determine whether there are foreign object events on overhead power lines. If an overhead power line foreign object event is detected, an early warning strategy is generated based on the type of the event and its location. If no such event is detected, the overhead power line foreign object detection video stream continues to be read.

11. Claim 11: The S1024 intelligent algorithm for personnel or large animal track encroachment in the S102 intelligent algorithm training function of claim 5, characterized in that... Analyzing collected video data on personnel or large animals encroaching on public transport routes and generating early warning strategies based on the analysis results includes the following steps: Using cameras to collect video streams for detecting encroachment on tracks by people or large animals; The algorithm for detecting human or large animal encroachment on the track is based on machine vision deep learning to analyze video data and determine whether there are any such events based on the analysis results. If a human or large animal encroachment incident occurs, an early warning strategy is generated based on the type of the incident and its detected location; if no human or large animal encroachment incident occurs, the video stream of the detected human or large animal encroachment incident continues to be read.

12. Claim 12: The S1025 intelligent smoke and fire alarm early warning algorithm in the S102 intelligent algorithm training function of claim 5, characterized in that... Analyzing the collected smoke and fire alarm video data and generating early warning strategies based on the analysis results includes the following steps: Use cameras to capture video streams of smoke and fire alarm detection; The algorithm for detecting smoke and fire alarms uses deep learning based on machine vision technology, combined with smoke sensors, to analyze video data and determine whether a smoke and fire alarm has occurred based on the analysis results. If a fire alarm event is detected, an early warning strategy is generated based on the type of fire alarm event and the location where the fire alarm event is detected; if no fire alarm event is detected, the fire alarm detection video stream is read again.

13. Claim 13: The S1026 intelligent early warning algorithm for platform personnel encroachment in the S102 intelligent algorithm training function described in claim 5, characterized in that... Analyzing the collected video data of people encroaching on platform space and generating early warning strategies based on the analysis results includes the following steps: Using cameras to collect video streams of people encroaching on platform limits while waiting for trains; The algorithm for detecting platform personnel encroachment on platform boundaries is based on the principle of deep learning to identify the inside and outside of the yellow line on the platform. The video data is analyzed, and the analysis results are used to determine whether there are any platform personnel encroachment on platform boundaries. If a platform passenger encroachment incident is detected, an early warning strategy is generated based on the type of the incident and its location. If no such incident is detected, the system continues to read the platform passenger encroachment detection video stream.

14. Claim 14: The S1027 intelligent early warning algorithm for platform personnel falls in the S102 intelligent algorithm training function described in claim 5, characterized in that... Analyzing the collected video data of people falling on the platform and generating early warning strategies based on the analysis results includes the following steps: Using cameras to collect video streams of people falling on the platform; The algorithm for detecting platform personnel falls is based on deep learning and machine vision. Video data is analyzed, and the results are used to determine whether there are any platform personnel fall incidents. If a platform personnel fall incident is detected, an early warning strategy is generated based on the type of platform personnel fall incident and the location where the incident was detected; if no platform personnel fall incident is detected, the platform personnel fall detection video stream is read again.

15. Claim 15: The S1028 debris flow intelligent early warning algorithm in the S102 intelligent algorithm training function of claim 5, characterized in that... Analyzing the collected debris flow video data and generating early warning strategies based on the analysis results includes the following steps: Using cameras to capture video streams of debris flow detection; A debris flow event detection algorithm based on deep learning methods for calculating debris flow flow rate and velocity is used to analyze video data, and the existence of debris flow events is determined based on the analysis results. If a debris flow event is detected, an early warning strategy is generated based on the debris flow event type and the location where the debris flow event is detected; if no debris flow event is detected, the debris flow detection video stream is read again.

16. Claim 16: The S1029 intelligent early warning algorithm for monitoring crowd congestion at the exit in the S102 intelligent algorithm training function of claim 5, characterized in that... Analyzing the collected video data on crowd congestion at the exit and generating early warning strategies based on the analysis results includes the following steps: Use cameras to collect video streams of crowd control at the station exit; An algorithm for detecting crowding events at station exits is used to analyze video data based on the theory of passenger flow queuing in video images to measure the congestion level in a specified area, and the analysis results are used to determine whether there are crowding events at the station exits. If a crowding event occurs at the exit, an early warning strategy is generated based on the type of crowding event and the location where the crowding event is detected; if no crowding event occurs at the exit, the video stream of the crowding detection at the exit continues to be read.

17. Claim 17: The types of intelligent algorithm training functions in S102 of the claim are not limited to those listed in claims 7-16 above. All artificial intelligence algorithms that are beneficial to the safety of railway line patrol can be applied by loading them onto the edge computer in the cloud platform S1 of the railway line intelligent patrol system.

Images