A subway intelligent perception method based on machine vision

Abstract

The present application belongs to the technical field of subway operation monitoring, and particularly relates to a subway intelligent sensing method based on machine vision. Through the machine vision sensing capability of the train itself, the present application realizes real-time monitoring and risk warning of the track state, effectively improving the safety and reliability of subway operation. In the case of communication interruption between the train and the ground, the on-board sensing unit can be quickly activated and take over the autonomous sensing task of the train. Through the pre-deployed on-board sensors, track environment data is collected, and in combination with real-time image capture and recognition technology, the track features are judged. Furthermore, through measures such as dynamically adjusting the safety distance and real-time monitoring of the turnout switching state, the autonomous decision-making capability of the train in complex environments is further enhanced. Not only does the present application solve the information loss problem of traditional monitoring systems in the case of communication interruption, but also provides strong protection for the safe operation of the subway.

Description

Technical Field

[0001] This invention belongs to the field of subway operation monitoring technology, specifically relating to a subway intelligent sensing method based on machine vision. Background Technology

[0002] With the acceleration of urbanization, the safety, reliability and efficiency of subways, as an important part of urban public transportation, are receiving increasing attention. Traditional subway monitoring systems mainly rely on communication between the ground control center and the train. However, under certain special circumstances, such as signal obstruction in tunnels or communication failures, communication between the train and the ground control center may be interrupted, causing the train to be unable to obtain track status information in a timely manner, which poses a safety hazard. Therefore, ensuring the safe operation of the subway under communication interruption has become an urgent problem to be solved.

[0003] In existing technologies, the ability of subways to perceive information during communication interruptions is usually improved by adding sensors or increasing communication redundancy. However, these methods often suffer from high costs, complex deployment, and difficult maintenance. In addition, increasing communication redundancy cannot completely solve the problem of information loss caused by communication interruptions. It still relies on the support of ground facilities and is difficult to provide continuous and reliable perception information in the event of long-term or large-scale communication interruptions. As a result, it is impossible to guarantee the autonomous perception and decision-making capabilities of trains without communication support. Based on this, this application provides a machine vision-based intelligent perception method for subways to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide a machine vision-based intelligent sensing method for subways, which can achieve real-time monitoring and risk warning of track status through the train's own visual perception without relying on ground communication facilities.

[0005] The specific technical solution adopted by this invention is as follows:

[0006] A machine vision-based intelligent sensing method for subways includes:

[0007] Track environment data is collected by pre-deployed onboard sensors, and the onboard sensing unit is activated after the communication interruption between the train and the ground has timed out, requesting the current track segment status information from the track signal equipment;

[0008] The target recognition range within the track area is determined based on the current road segment status information, and real-time image capture is performed on the target recognition range to obtain real-time image data;

[0009] Based on real-time image data, train track characteristics are identified, including straight tracks and bifurcated tracks.

[0010] On straight tracks, track occupancy detection is performed, and the safety distance is dynamically adjusted based on train speed and braking distance.

[0011] Under bifurcated tracks, identify the position of the turnout, determine the direction of train travel based on the turnout position, and confirm the train travel path by combining the turnout switching status with the train operation plan.

[0012] In a preferred embodiment, the vehicle-mounted sensors include a lidar, an infrared camera, and a millimeter-wave radar. The lidar is used to acquire three-dimensional point cloud data of the track, the infrared camera is used to acquire images of the track environment in low-light conditions, and the millimeter-wave radar is used to detect obstacle distance information. After the data from the millimeter-wave radar and lidar are acquired, data fusion processing is performed to obtain obstacle position and shape information. Combined with the environmental images acquired by the infrared camera, image enhancement processing is performed to obtain a multi-dimensional perception image of the track environment.

[0013] In a preferred embodiment, the step of activating the onboard sensing unit and requesting current track segment status information from the track signaling equipment after the train-to-ground communication interruption timeout includes:

[0014] The communication link quality parameters between the train and the ground control center are acquired in real time. These parameters include signal strength, signal-to-noise ratio, and signal feedback delay.

[0015] The quality parameters of each communication link are fused and calculated to obtain the communication link quality score;

[0016] When the communication link quality score is higher than or equal to the predetermined threshold, the communication link is determined to be normal, and the current communication mode is maintained for information exchange between the train and the ground.

[0017] When the communication link quality score is lower than the predetermined threshold, it indicates an anomaly between the train and the ground communication link. A continuous monitoring window is constructed starting from the anomaly calibration node. If the communication link quality parameter within the continuous monitoring window is continuously lower than the predetermined threshold, it is determined that the communication is interrupted and the activation command of the on-board sensing unit is triggered.

[0018] Among them, after the on-board sensing unit is activated, it takes over the train's autonomous sensing task and calls the locally stored track topology data and historical operation information as the basis for environmental modeling.

[0019] Track boundaries are extracted using lidar point cloud data and infrared images, and obstacle distance parameters fed back by millimeter-wave radar are combined to construct a three-dimensional spatial situation map. Train operation is then conducted with the assistance of this three-dimensional spatial situation map.

[0020] In a preferred embodiment, the step of determining the target identification range within the track area based on the current road segment status information includes:

[0021] Extract the track topology identifier from the current road segment status information, and match the corresponding sensing area range parameters through the track topology identifier. The sensing area range parameters include the track curvature radius, slope information, and clearance dimensions.

[0022] Based on the sensing area range parameters, the cooperative scanning angle of lidar and millimeter-wave radar and the imaging field of view of infrared camera are determined.

[0023] The scanning density of the lidar is adjusted according to the radius of curvature of the track. The smaller the radius of curvature of the track, the higher the scanning density of the lidar, so as to ensure the integrity of the point cloud data in the curved area.

[0024] The elevation angle of the millimeter-wave radar is adjusted by combining the slope information so that the radar beam is focused in front of the track.

[0025] Under the constraint of the limit size, the imaging field of view of the infrared camera is limited to a fixed area on both sides of the track.

[0026] In a preferred embodiment, the step of identifying train track features based on real-time image data includes:

[0027] Real-time image data is preprocessed to eliminate image noise and enhance contrast, resulting in a standardized orbital scene image;

[0028] The edge contours of the track structure are separated from the standardized track scene diagram, and track morphology parameters are generated through spatial geometric relationships;

[0029] The track morphology parameters are matched and compared with a preset standard track feature library to identify whether the track feature type of the train's current travel segment is a straight track or a bifurcated track.

[0030] In a preferred embodiment, the step of performing track occupancy detection and dynamically adjusting the safety distance in conjunction with train speed and braking distance includes:

[0031] The presence of foreign objects or obstructions within the track area is identified by combining lidar point cloud data with infrared images.

[0032] If it does not exist, the train continues to run at the set speed;

[0033] If present, the real-time distance between the track occupancy point and the front of the train is collected.

[0034] Real-time train speed information is collected, and the safe braking distance is calculated by combining the train braking performance parameters;

[0035] A dynamic safe distance threshold is generated based on the safe braking distance and the real-time distance.

[0036] When the real-time distance is less than or equal to the safe distance threshold, a warning signal is issued simultaneously, and a graded braking command is generated to control the train to decelerate or brake urgently.

[0037] If the track occupancy is still not cleared after the train has slowed down and traveled to a safe braking distance, emergency braking shall be applied immediately.

[0038] If the track occupancy is cleared before the train slows down and travels to a safe braking distance, the warning signal will be automatically canceled and the train will resume operation at the set speed.

[0039] In a preferred embodiment, the step of determining the train's direction of travel based on the switch position includes:

[0040] Obtain the geometric characteristics of the turnout point rail and the stock rail under the bifurcation track, and determine the track bifurcation angle based on the extension direction of the turnout point rail.

[0041] Identify the target track branch that the train is allowed to enter based on the angle between the turnout bifurcation angle and the train's current direction of travel;

[0042] Match and compare the target trajectory branch with the planned path;

[0043] If a target track branch exists that matches the planned path, the train will enter the corresponding target track branch and continue running.

[0044] If there is no target track branch that matches the planned path, it indicates that the turnout position is abnormal or there is a turnout switch.

[0045] When the switch position is abnormal, the fault alarm mechanism is automatically triggered and a braking command is sent to the train;

[0046] During turnout switching, the displacement trajectory of the turnout switch rail is monitored in real time. When the displacement trajectory of the switch rail matches the preset switching trajectory, the turnout switching is determined to be complete, and the matching and identification of the target track branch is re-executed.

[0047] In a preferred embodiment, after the onboard sensing unit is activated, the train continuously sends request signals to the ground control center and collects the communication link quality score of the request signals in real time.

[0048] If the communication link quality score of the requested signal exceeds a predetermined threshold, and the duration of exceeding the predetermined threshold is greater than a predetermined time threshold, the on-board sensing unit is shut down, and the connection between the train and the ground control center is re-established.

[0049] The present invention also provides a machine vision-based intelligent sensing system for subways, which uses the above-mentioned machine vision-based intelligent sensing method for subways, including:

[0050] The operation monitoring module is used to collect track environment data through pre-deployed on-board sensors, and after the train-to-ground communication interruption timeout, it activates the on-board sensing unit to request the current track segment status information from the track signal equipment.

[0051] The target range recognition module is used to determine the target recognition range within the track area based on the current road segment status information, and to capture real-time images of the target recognition range to obtain real-time image data.

[0052] The track feature recognition module is used to identify the track features of train operation based on real-time image data. The track features of train operation include straight tracks and bifurcated tracks.

[0053] The straight track monitoring module is used to perform track occupancy detection on straight tracks and dynamically adjust the safety distance based on train speed and braking distance.

[0054] The turnout monitoring module is used to identify the position of the turnout on the branched track, determine the direction of train travel based on the turnout position, and confirm the train travel path by combining the turnout switching status with the train operation plan.

[0055] And, an electronic device, the electronic device comprising:

[0056] At least one processor;

[0057] and a memory communicatively connected to the at least one processor;

[0058] The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the aforementioned machine vision-based intelligent sensing method for subways.

[0059] The technical effects achieved by this invention are as follows:

[0060] This invention utilizes the train's own machine vision perception capabilities to achieve real-time monitoring and risk warning of track conditions, effectively improving the safety and reliability of subway operation. In the event of a communication interruption between the train and the ground, the onboard perception unit can be quickly activated and take over the train's autonomous perception tasks. By collecting track environment data through pre-deployed onboard sensors and combining real-time image capture and recognition technology, it can determine track characteristics. Furthermore, by dynamically adjusting the safety distance and monitoring the switch switching status in real time, it further enhances the train's autonomous decision-making ability in complex environments. This not only solves the problem of information loss in traditional monitoring systems when communication is interrupted, but also provides strong protection for the safe operation of the subway. Attached Figure Description

[0061] Figure 1This is a schematic diagram of the method flow of the present invention;

[0062] Figure 2 This is a schematic diagram of the system modules of the present invention;

[0063] Figure 3 This is a schematic diagram of the electronic device structure of the present invention. Detailed Implementation

[0064] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0065] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0066] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0067] Please see Figure 1 As shown, this invention provides a machine vision-based intelligent sensing method for subways, comprising:

[0068] S1. Collect track environment data through pre-deployed onboard sensors, and after the train-to-ground communication interruption timeout, activate the onboard sensing unit to request the current track segment status information from the track signal equipment;

[0069] In step S1, the subway, as one of the main modes of transportation for people's daily travel, is of paramount importance in terms of safety and efficiency. Because the subway operates in underground tunnels, it is often accompanied by unstable communication signals, especially in deep tunnels or special terrain areas. Onboard sensors may be unable to communicate with the ground control system in a timely manner due to signal shielding, which may cause corresponding safety hazards. In this embodiment, the pre-deployed onboard sensors first collect track environment data. In the event of a communication interruption between the train and the ground, the onboard sensing unit is activated in a timely manner to request the current track status information from the track signal equipment, so as to ensure that the train can autonomously perceive the surrounding environment and make safety decisions accordingly. The onboard sensors include lidar, infrared camera and millimeter-wave radar. LiDAR is used to acquire three-dimensional point cloud data of the track, infrared camera is used to acquire track environment images in low light environment, and millimeter-wave radar is used to detect obstacle distance information. After the data is collected by millimeter-wave radar and lidar, data fusion processing is performed to obtain obstacle position and shape information. Combined with the environmental image collected by infrared camera, image enhancement processing is performed to obtain a multi-dimensional perception image of the track environment.

[0070] Specifically, the onboard sensors mainly include lidar, infrared cameras, and millimeter-wave radar, forming a machine vision multimodal perception system. This system can collect and analyze track environment information in real time, ensuring the train's perception of its surroundings even during communication interruptions, thereby improving train operation safety. The main function of lidar is to collect three-dimensional point cloud data of the track for subsequent track structure modeling. Infrared cameras are used to capture clear track environment images under low-light conditions, such as inside tunnels or at night, effectively acquiring visual information about the track's surroundings. Millimeter-wave radar is responsible for detecting the distance and speed information of surrounding obstacles, ensuring the train can accurately perceive the situation ahead even in complex environments. Subsequently, the data collected by millimeter-wave radar and lidar will be fused. The fusion method can use the Kalman filter algorithm to fuse the two sets of data to obtain the location and shape information of obstacles. At the same time, combined with the environmental images collected by the infrared camera, image enhancement algorithms are used to denoise and improve the contrast of the images. Finally, the multi-dimensional data is integrated into a multi-dimensional perception image of the track environment, providing corresponding environmental perception support for the train's continued operation and ensuring that the train can accurately judge the surrounding environment even during communication interruptions.

[0071] Secondly, after the communication interruption between the train and the ground times out, the steps to activate the onboard sensing unit and request the current track segment status information from the track signaling equipment include:

[0072] The communication link quality parameters between the train and the ground control center are acquired in real time. These parameters include signal strength, signal-to-noise ratio, and signal feedback delay.

[0073] The quality parameters of each communication link are fused and calculated to obtain the communication link quality score;

[0074] When the communication link quality score is higher than or equal to the predetermined threshold, the communication link is determined to be normal, and the current communication mode is maintained for information exchange between the train and the ground.

[0075] When the communication link quality score is lower than the predetermined threshold, it indicates an anomaly between the train and the ground communication link. A continuous monitoring window is constructed starting from the anomaly calibration node. If the communication link quality parameter within the continuous monitoring window is continuously lower than the predetermined threshold, it is determined that the communication is interrupted and the activation command of the on-board sensing unit is triggered.

[0076] Among them, after the on-board sensing unit is activated, it takes over the train's autonomous sensing task and calls the locally stored track topology data and historical operation information as the basis for environmental modeling.

[0077] Track boundaries are extracted using lidar point cloud data and infrared images, and obstacle distance parameters fed back by millimeter-wave radar are combined to construct a three-dimensional spatial situation map. Train operation is then carried out with the assistance of the three-dimensional spatial situation map.

[0078] In this implementation, when determining the communication quality between the train and the ground control center, the communication link quality parameters are monitored in real time to ensure that the train can respond promptly to changes in the communication status. The communication link quality parameters mainly consist of three parts: signal strength, signal-to-noise ratio (SNR), and signal feedback delay. Signal strength reflects the received signal power and is one of the important indicators for evaluating communication link quality; higher signal strength indicates better communication link quality. SNR is the ratio of signal power to noise power; a higher SNR indicates less interference during signal transmission and more stable communication quality. Signal feedback delay is the time required for the signal to travel from transmission to reception, reflecting the response speed of the communication link; a lower delay indicates better real-time performance. Then, the three parameters—signal strength, SNR, and signal feedback delay—are fused. The fusion calculation can be weighted fusion or exponential decay fusion, depending on the actual needs. This results in a communication link quality score that reflects the overall communication quality. A communication link quality score greater than or equal to... When the communication link quality score is below the predetermined threshold, it indicates that the current communication link is in normal condition and can continue to maintain the existing communication mode to ensure information exchange between the train and the ground control center. When the communication link quality score is below the predetermined threshold, it indicates that the communication link between the train and the ground control center is abnormal. At this time, a continuous monitoring window is built starting from the abnormality calibration node. If the communication link quality parameter continues to be below the predetermined threshold within the continuous monitoring window, it is determined that the communication link has been interrupted, which triggers the activation command of the on-board sensing unit. This ensures that the train still has the ability to operate autonomously when it loses ground communication support. After activation, the on-board sensing unit takes over the train's environmental perception task. It builds a basic environmental model by calling the locally stored track topology data and historical operation information. At the same time, it integrates lidar point cloud data, infrared image information, and obstacle distance parameters fed back by millimeter-wave radar to perform multi-source heterogeneous data fusion processing, further improving the accuracy of environmental modeling. This ensures that the train can still accurately identify changes in the surrounding environment and make corresponding operation decisions when communication is interrupted.

[0079] S2. Determine the target recognition range within the track area based on the current road segment status information, and capture real-time images of the target recognition range to obtain real-time image data;

[0080] In step S2, after the onboard sensing unit is activated, it requests the current track segment status information from the track signaling equipment and determines the target recognition range within the track area based on this information. This target recognition range typically covers the track and its surrounding area within a certain distance in front of the train to ensure comprehensive capture of obstacles or abnormal situations that may affect train operation. Simultaneously, real-time image data within the target recognition range is captured. The step of determining the target recognition range within the track area based on the current track segment status information includes:

[0081] Extract the track topology identifier from the current road segment status information, and match the corresponding sensing area range parameters through the track topology identifier. The sensing area range parameters include the track curvature radius, slope information, and clearance dimensions.

[0082] Based on the sensing area range parameters, the cooperative scanning angle of lidar and millimeter-wave radar and the imaging field of view of infrared camera are determined.

[0083] The scanning density of the lidar is adjusted according to the radius of curvature of the track. The smaller the radius of curvature of the track, the higher the scanning density of the lidar, so as to ensure the integrity of the point cloud data in the curved area.

[0084] The elevation angle of the millimeter-wave radar is adjusted by combining the slope information so that the radar beam is focused in front of the track.

[0085] Under the constraint of the limit size, the imaging field of view of the infrared camera is limited to a fixed area on both sides of the track;

[0086] Specifically, when determining the target recognition range, the first step is to extract the track topology identifier from the current road segment status information. The track topology identifier contains information such as the track's geometric features and operational restrictions. Through the track topology, pre-set sensing area parameters can be quickly matched. These parameters describe the required monitoring range for different road segments, covering track curvature radius, slope information, and clearance dimensions. Based on these sensing area parameters, the operating parameters of the onboard sensors can be intelligently adjusted. For example, the scanning angle and density of the LiDAR will be dynamically adjusted according to changes in the track curvature radius, ensuring detection even in complex terrain such as curves. Once the corresponding point cloud data is acquired, a scanning density algorithm for the lidar can be used to dynamically adjust the radius of curvature of the track. The detection elevation angle of the millimeter-wave radar will also be optimized in conjunction with the slope information to ensure that the radar beam can always be focused on the front of the track, accurately detecting the distance and speed information of obstacles. Specifically, a commonly used millimeter-wave radar elevation angle optimization algorithm can be used. The imaging field of view of the infrared camera will also be constrained by the limit size to ensure that the image it captures is focused on a fixed area on both sides of the track, avoiding interference from irrelevant information and improving the efficiency of image processing, thereby achieving high-precision perception of the track environment and identification of abnormal targets.

[0087] S3. Identify the characteristics of the train's running track based on real-time image data, including straight tracks and bifurcated tracks;

[0088] In step S3, after the real-time image data is output, a train track feature identification operation is performed to determine whether the current track type is a straight track or a bifurcated track. The step of identifying train track features based on real-time image data includes:

[0089] Real-time image data is preprocessed to eliminate image noise and enhance contrast, resulting in a standardized orbital scene image;

[0090] The edge contours of the track structure are separated from the standardized track scene diagram, and track morphology parameters are generated through spatial geometric relationships;

[0091] The track morphology parameters are matched and compared with a preset standard track feature library to identify whether the track feature type of the current train travels is a straight track or a bifurcated track.

[0092] When determining the characteristics of a train track, the real-time image data is first preprocessed. Preprocessing methods include filtering and noise reduction, and contrast enhancement, to obtain a clear, standardized track scene image. Then, the edge contours of the track structure are separated from the standardized track scene image. This is achieved through edge detection, which highlights edge information in the real-time image data and ignores irrelevant areas. Common edge detection methods include the Sobel operator and the Canny operator, and the specific method chosen depends on the actual image characteristics and processing requirements. After edge detection, the edge contours of the track structure are obtained. Then, based on these edge contours, spatial geometric relationships are used to... The morphological parameters of the track are calculated, such as the radius of curvature, gauge width, and deflection angle. The specific calculation method can be obtained by fitting and analyzing the geometric characteristics of the track edge profile. For example, for a straight track, its radius of curvature is theoretically infinite, the gauge width remains constant, and the deflection angle is zero. However, for a bifurcated track, the track branches will appear at specific locations, at which point the gauge width will change and a significant deflection angle will be generated. By comparing the calculated morphological parameters with a preset standard track feature library, it can be determined whether the current train travels on a straight track or a bifurcated track, thus providing reliable data support for subsequent operation control and safety warning.

[0093] S4. On straight tracks, track occupancy detection is performed, and the safety distance is dynamically adjusted based on train speed and braking distance.

[0094] In step S4, the train maintains a predetermined speed on a straight track. However, it is necessary to monitor whether there are construction sites or obstacles obstructing the path ahead. Real-time detection of track occupancy is required, and the safe distance between the train and obstacles is dynamically calculated and adjusted based on the train's current speed and braking performance parameters. This ensures the train can brake promptly in emergencies to avoid collisions. The steps of detecting track occupancy and dynamically adjusting the safe distance based on train speed and braking distance include:

[0095] The presence of foreign objects or obstructions within the track area is identified by combining lidar point cloud data with infrared images.

[0096] If it does not exist, the train continues to run at the set speed;

[0097] If present, the real-time distance between the track occupancy point and the front of the train is collected.

[0098] Real-time train speed information is collected, and the safe braking distance is calculated by combining the train braking performance parameters;

[0099] A dynamic safe distance threshold is generated based on the safe braking distance and the real-time distance.

[0100] When the real-time distance is less than or equal to the safe distance threshold, a warning signal is issued simultaneously, and a graded braking command is generated to control the train to decelerate or brake urgently.

[0101] If the track occupancy is still not cleared after the train has slowed down and traveled to a safe braking distance, emergency braking shall be applied immediately.

[0102] If the track occupancy is cleared before the train slows down and travels to a safe braking distance, the warning signal will be automatically canceled and the train will resume operation at the set speed.

[0103] Specifically, when performing track occupancy detection on straight tracks, the track area is primarily monitored using lidar point cloud data and infrared images. Lidar emits laser beams and receives reflected signals to generate three-dimensional point cloud data of the track area, reflecting the shape of the track surface and any potential obstacles. Infrared cameras, on the other hand, can capture image information of the track and its surroundings under low-light conditions, further enhancing the ability to perceive track occupancy. When both lidar and infrared images confirm that there are no foreign objects or occupancy within the track area, the train can continue safely at a set speed. Once a foreign object or occupancy is detected on the track, the distance between the occupancy point and the front of the train is collected in real time, along with the train's operating speed. Combined with the train's braking performance parameters, a safe braking distance is dynamically calculated. In the formula, Indicates the safe braking distance. Indicates the train's speed. This indicates that braking causes the train to decelerate. This indicates the braking system response time. The safe braking distance refers to the shortest distance a train needs to decelerate from its current speed to a complete stop in an emergency braking situation. After obtaining the safe braking distance, a dynamic safe distance threshold is generated based on it and the real-time distance. The safe distance threshold = safety factor * safe braking distance. The safety factor is a preset value greater than 1 based on the actual operating environment and safety requirements of the train, ensuring that the train has sufficient safety margin for braking in an emergency. When the real-time distance is less than or equal to the safe distance threshold, a warning signal is issued simultaneously to alert the train driver, and a graded braking command is automatically generated. The graded braking command gradually controls the train to decelerate according to the degree of reduction in the real-time distance until emergency braking. If the track occupancy is not cleared after the train has decelerated and advanced to the safe braking distance, emergency braking will be triggered immediately to ensure the train can stop smoothly and avoid a collision. If the track occupancy is cleared before the train decelerates and advances to the safe braking distance, the warning signal will be automatically canceled, and the train will be gradually restored to the set speed to ensure safe operation of the train on a straight track.

[0104] S5. Under the bifurcated track, identify the position of the turnout, determine the direction of train travel based on the position of the turnout, and confirm the train travel path by combining the turnout switching status and the train operation plan.

[0105] In step S5, under the branched track, the train's direction of travel is determined by identifying the switch position to ascertain whether the train is traveling along the predetermined path. If the switch position is found to be inconsistent with the train's operating plan, a path anomaly warning will be immediately issued, and the train driver will be notified to take appropriate measures to intervene and prevent the train from occupying the normal travel path of other trains. The step of determining the train's direction of travel based on the switch position includes:

[0106] Obtain the geometric characteristics of the turnout point rail and the stock rail under the bifurcation track, and determine the track bifurcation angle based on the extension direction of the turnout point rail.

[0107] Identify the target track branch that the train is allowed to enter based on the angle between the turnout bifurcation angle and the train's current direction of travel;

[0108] Match and compare the target trajectory branch with the planned path;

[0109] If a target track branch exists that matches the planned path, the train will enter the corresponding target track branch and continue running.

[0110] If there is no target track branch that matches the planned path, it indicates that the turnout position is abnormal or there is a turnout switch.

[0111] When the switch position is abnormal, the fault alarm mechanism is automatically triggered and a braking command is sent to the train;

[0112] During turnout switching, the displacement trajectory of the turnout switch rail is monitored in real time. When the switch rail displacement trajectory matches the preset switching trajectory, the turnout switching is determined to be complete, and the matching and identification of the target track branch is re-executed.

[0113] Under bifurcated tracks, the geometric features of the corresponding switch rail and stock rail are first collected. These features include the length, shape, and position of the switch rail relative to the stock rail. Based on these geometric features, the bifurcation angle is calculated. The bifurcation angle is the angle between the extension direction of the switch rail and the current direction of train travel. Specifically, it is determined by analyzing the vector angle between the extension direction of the switch rail and the direction of train travel. After obtaining the bifurcation angle, it is compared with the current direction of train travel to identify the target track branch that the train is allowed to enter. The identified target track branch is then matched against the predetermined path in the train's travel plan to confirm whether the train is traveling along the predetermined path. If the match is successful (i.e., a target track branch exists that matches the planned path), the train will enter the corresponding target track branch to continue its journey. If the match fails (i.e., no target track branch exists), the train will enter the corresponding target track branch to continue its journey. If the target track branch matches the planned path, it will determine whether there is an abnormal turnout position. If there is an abnormal turnout position, it may be due to a turnout malfunction or human error. In this case, a fault alarm mechanism will be triggered, and a braking command will be sent to the train to ensure that the train stops at a safe position to avoid dangerous situations such as derailment or collision. At the same time, the fault alarm information will also be sent to the ground control center in a timely manner after communication is restored so that maintenance personnel can quickly locate and repair the fault. If it is determined to be a turnout switch, that is, the turnout is moving from its current position to another position to meet the train's driving needs, the displacement trajectory of the turnout switch rail will be monitored in real time to ensure that the switch rail can move according to the preset switching trajectory. When the displacement trajectory of the switch rail completely matches the preset switching trajectory, the turnout switch is determined to be complete, and then the target track branch matching and identification operation will be re-executed to ensure that the train can accurately enter the predetermined target track branch.

[0114] In addition, after the onboard sensing unit is activated, the train continuously sends request signals to the ground control center and collects the communication link quality score of the request signals in real time.

[0115] If the communication link quality score of the requested signal exceeds a predetermined threshold, and the duration of exceeding the predetermined threshold is greater than a predetermined time threshold, the onboard sensing unit is shut down, and the connection between the train and the ground control center is re-established.

[0116] In this implementation, after the communication link between the train and the ground control center is interrupted, the train will continue to send request signals to the ground control center and re-establish the connection as soon as the communication link is restored to ensure the timely transmission of operation control commands and the safety of train operation. The specific process is the same as the communication link quality scoring mechanism in step S1 above, thereby realizing dynamic closed-loop management. After the train and the ground control center re-establish the connection, the train's operation data during the communication interruption will also be uploaded synchronously so that the ground control center can fully grasp the status changes and operation trajectory of the train during the communication interruption.

[0117] Please see Figure 2 A machine vision-based intelligent sensing system for subways, using the aforementioned machine vision-based intelligent sensing method for subways, includes:

[0118] The operation monitoring module is used to collect track environment data through pre-deployed on-board sensors, and after the train-to-ground communication interruption timeout, it activates the on-board sensing unit to request the current track segment status information from the track signal equipment.

[0119] The target range recognition module is used to determine the target recognition range within the track area based on the current road segment status information, and to capture real-time images of the target recognition range to obtain real-time image data.

[0120] The track feature recognition module is used to identify the track features of train operation based on real-time image data. The track features of train operation include straight tracks and bifurcated tracks.

[0121] The straight track monitoring module is used to perform track occupancy detection on straight tracks and dynamically adjust the safety distance based on train speed and braking distance.

[0122] The turnout monitoring module is used to identify the position of the turnout under the branching track, determine the direction of train travel based on the position of the turnout, and confirm the train travel path by combining the turnout switching status and the train operation plan.

[0123] The execution flow of the aforementioned perception system is consistent with that of the machine vision-based intelligent perception method for subways, and will not be repeated here.

[0124] Please see Figure 3 An electronic device, comprising:

[0125] At least one processor;

[0126] and memory that is communicatively connected to at least one processor;

[0127] The memory stores a computer program that can be executed by at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the aforementioned machine vision-based intelligent sensing method for subways.

[0128] The processor of the aforementioned electronic device can be a central processing unit (CPU), a graphics processing unit (GPU), or an application-specific integrated circuit (ASIC). The memory can be a non-volatile storage medium such as random access memory (RAM), read-only memory (ROM), or flash memory. The electronic device may also include an arithmetic logic unit (ALU) and a controller, which are used to cooperate with the processor to complete data operations and instruction control, as well as input devices and output devices. The input devices can be a touch screen, a mouse, or a keyboard, and the output devices can be a display screen or a speaker, which are used for human-computer interaction and information display.

[0129] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0130] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A machine vision-based intelligent sensing method for subways, characterized in that: include: Track environment data is collected by pre-deployed onboard sensors, and the onboard sensing unit is activated after the communication interruption between the train and the ground has timed out, requesting the current track segment status information from the track signal equipment; The target recognition range within the track area is determined based on the current road segment status information, and real-time image capture is performed on the target recognition range to obtain real-time image data; Based on real-time image data, train track characteristics are identified, including straight tracks and bifurcated tracks. On straight tracks, track occupancy detection is performed, and the safety distance is dynamically adjusted based on train speed and braking distance. Under bifurcated tracks, identify the position of the turnout, determine the direction of train travel based on the position of the turnout, and confirm the train travel path by combining the turnout switching status with the train operation plan; The step of determining the target identification range within the track area based on the current road segment status information includes: Extract the track topology identifier from the current road segment status information, and match the corresponding sensing area range parameters through the track topology identifier. The sensing area range parameters include the track curvature radius, slope information, and clearance dimensions. Based on the sensing area range parameters, the cooperative scanning angle of lidar and millimeter-wave radar and the imaging field of view of infrared camera are determined. The scanning density of the lidar is adjusted according to the radius of curvature of the track. The smaller the radius of curvature of the track, the higher the scanning density of the lidar, so as to ensure the integrity of the point cloud data in the curved area. The elevation angle of the millimeter-wave radar is adjusted by combining the slope information so that the radar beam is focused in front of the track. Under the constraint of the limit size, the imaging field of view of the infrared camera is limited to a fixed area on both sides of the track.

2. The intelligent sensing method for subways based on machine vision according to claim 1, characterized in that: The vehicle-mounted sensors include lidar, infrared camera, and millimeter-wave radar. LiDAR is used to acquire three-dimensional point cloud data of the track, infrared camera is used to acquire track environment images in low-light conditions, and millimeter-wave radar is used to detect obstacle distance information. After the data from the millimeter-wave radar and lidar are acquired, data fusion processing is performed to obtain obstacle position and shape information. Combined with the environmental images acquired by the infrared camera, image enhancement processing is performed to obtain a multi-dimensional perception image of the track environment.

3. The intelligent sensing method for subways based on machine vision according to claim 2, characterized in that: The step of activating the onboard sensing unit and requesting current track segment status information from the track signaling equipment after the train-to-ground communication interruption timeout includes: The communication link quality parameters between the train and the ground control center are acquired in real time. These parameters include signal strength, signal-to-noise ratio, and signal feedback delay. The quality parameters of each communication link are fused and calculated to obtain the communication link quality score; When the communication link quality score is higher than or equal to the predetermined threshold, the communication link is determined to be normal, and the current communication mode is maintained for information exchange between the train and the ground. When the communication link quality score is lower than the predetermined threshold, it indicates an anomaly between the train and the ground communication link. A continuous monitoring window is constructed starting from the anomaly calibration node. If the communication link quality parameter within the continuous monitoring window is continuously lower than the predetermined threshold, it is determined that the communication is interrupted and the activation command of the on-board sensing unit is triggered. Among them, after the on-board sensing unit is activated, it takes over the train's autonomous sensing task and calls the locally stored track topology data and historical operation information as the basis for environmental modeling. Track boundaries are extracted using lidar point cloud data and infrared images, and obstacle distance parameters fed back by millimeter-wave radar are combined to construct a three-dimensional spatial situation map. Train operation is then conducted with the assistance of this three-dimensional spatial situation map.

4. The intelligent sensing method for subways based on machine vision according to claim 3, characterized in that: The step of identifying train track features based on real-time image data includes: Real-time image data is preprocessed to eliminate image noise and enhance contrast, resulting in a standardized orbital scene image; The edge contours of the track structure are separated from the standardized track scene diagram, and track morphology parameters are generated through spatial geometric relationships; The track morphology parameters are matched and compared with a preset standard track feature library to identify whether the track feature type of the train's current travel segment is a straight track or a bifurcated track.

5. The intelligent sensing method for subways based on machine vision according to claim 1, characterized in that: The steps of performing track occupancy detection and dynamically adjusting the safety distance in conjunction with train speed and braking distance include: The presence of foreign objects or obstructions within the track area is identified by combining lidar point cloud data with infrared images. If it does not exist, the train continues to run at the set speed; If present, the real-time distance between the track occupancy point and the front of the train is collected. Real-time train speed information is collected, and the safe braking distance is calculated by combining the train braking performance parameters; A dynamic safe distance threshold is generated based on the safe braking distance and the real-time distance. When the real-time distance is less than or equal to the safe distance threshold, a warning signal is issued simultaneously, and a graded braking command is generated to control the train to decelerate or brake urgently. If the track occupancy is still not cleared after the train has slowed down and traveled to a safe braking distance, emergency braking shall be applied immediately. If the track occupancy is cleared before the train slows down and travels to a safe braking distance, the warning signal will be automatically canceled and the train will resume operation at the set speed.

6. The intelligent sensing method for subways based on machine vision according to claim 1, characterized in that: The step of determining the train's direction of travel based on the switch position includes: Obtain the geometric characteristics of the turnout point rail and the stock rail under the bifurcation track, and determine the track bifurcation angle based on the extension direction of the turnout point rail. Identify the target track branch that the train is allowed to enter based on the angle between the turnout bifurcation angle and the train's current direction of travel; Match and compare the target trajectory branch with the planned path; If a target track branch exists that matches the planned path, the train will enter the corresponding target track branch and continue running. If there is no target track branch that matches the planned path, it indicates that the turnout position is abnormal or there is a turnout switch. When the switch position is abnormal, the fault alarm mechanism is automatically triggered and a braking command is sent to the train; During turnout switching, the displacement trajectory of the turnout switch rail is monitored in real time. When the displacement trajectory of the switch rail matches the preset switching trajectory, the turnout switching is determined to be complete, and the matching and identification of the target track branch is re-executed.

7. The intelligent sensing method for subways based on machine vision according to claim 1, characterized in that: After the onboard sensing unit is activated, the train continuously sends request signals to the ground control center and collects the communication link quality score of the request signals in real time. If the communication link quality score of the requested signal exceeds a predetermined threshold, and the duration of exceeding the predetermined threshold is greater than a predetermined time threshold, the on-board sensing unit is shut down, and the connection between the train and the ground control center is re-established.

8. A machine vision-based intelligent sensing system for subways, characterized in that: The subway intelligent sensing method based on machine vision according to any one of claims 1 to 7 includes: The operation monitoring module is used to collect track environment data through pre-deployed on-board sensors, and after the train-to-ground communication interruption timeout, it activates the on-board sensing unit to request the current track segment status information from the track signal equipment. The target range recognition module is used to determine the target recognition range within the track area based on the current road segment status information, and to capture real-time images of the target recognition range to obtain real-time image data. The track feature recognition module is used to identify the track features of train operation based on real-time image data. The track features of train operation include straight tracks and bifurcated tracks. The straight track monitoring module is used to perform track occupancy detection on straight tracks and dynamically adjust the safety distance based on train speed and braking distance. The turnout monitoring module is used to identify the position of the turnout on the branched track, determine the direction of train travel based on the turnout position, and confirm the train travel path by combining the turnout switching status with the train operation plan.

9. An electronic device, characterized in that: The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the machine vision-based intelligent sensing method for subways as described in any one of claims 1 to 7.

Images