Track foreign matter detection system based on machine vision

Abstract

The invention belongs to the technical field of track supervision, and particularly relates to a track foreign matter detection system based on machine vision, which comprises a high-definition camera array unit, a high-performance computing unit, a track potential risk decision-making unit and a remote monitoring center, according to the invention, the high-definition camera array unit monitors the track in all directions, the high-performance computing unit judges whether the dangerous foreign matters exist on the track based on the track, and when the dangerous foreign matters are detected, track maintenance personnel are arranged to go to process the dangerous foreign matters, so that the intelligent and automatic monitoring of the track foreign matters is realized; the railway safety monitoring efficiency and accuracy are remarkably improved, the burden of manual inspection is effectively relieved, the potential risk of track management is subjected to decision evaluation analysis through the track potential risk decision unit, track management personnel are reminded to strengthen follow-up supervision on the track when a track potential high-risk signal is generated, and the safety of the railway is improved. And the rail operation safety is further ensured.

Description

Technical Field

[0001] The invention relates to the technical field of track supervision, and in particular to a track foreign body detection system based on machine vision. Background Art

[0002] Tracks are the routes paved with strips of steel or other materials for trains, trams and other vehicles to travel on. They provide a stable and continuous travel path for vehicles and are an important part of the modern transportation system. Foreign objects such as stones, metal fragments, and garbage on the tracks pose a serious threat to train operation and can easily lead to train derailment, collision and other accidents. Traditional manual inspection methods are inefficient and prone to missed inspections, making it difficult to identify and remove dangerous foreign objects on the track in a timely manner. Currently, foreign object detection on the track is generally carried out through machine vision technology, but in actual application, it is impossible to accurately assess the potential risks of track management while realizing intelligent and automated monitoring of foreign objects on the track. It is difficult for track management personnel to strengthen subsequent supervision of the track in a timely manner, which is not conducive to ensuring the safety of track operation. In view of the above technical defects, a solution is now proposed. Summary of the invention

[0003] The purpose of the present invention is to provide a track foreign object detection system based on machine vision, which solves the problem that the existing technology is unable to accurately evaluate the potential risks of track management while realizing intelligent and automated monitoring of track foreign objects, and it is difficult for track management personnel to strengthen subsequent supervision of the track in a timely manner, which is not conducive to ensuring the safety of track operation.

[0004] To achieve the above object, the present invention provides the following technical solutions: A track foreign body detection system based on machine vision, comprising a high-definition camera array unit, a high-performance computing unit, a track potential risk decision unit and a remote monitoring center; the high-definition camera array unit is composed of a plurality of groups of high-definition cameras, which are installed on both sides or above the track, and adopt a high-resolution and wide-viewing angle design to ensure that the entire track area is covered, and has a night infrared imaging capability, and the track is fully monitored through the high-definition camera array unit, and the collected track images are sent to the high-performance computing unit in real time; The high-performance computing unit processes and analyzes the received track images to determine whether there are dangerous foreign objects on the track. When dangerous foreign objects are detected, the early warning mechanism is triggered to generate corresponding early warning information, and the early warning information and the location of the dangerous foreign objects are sent to the remote monitoring center. When the remote monitoring center receives the early warning information, it arranges track maintenance personnel to go and handle the dangerous foreign objects. The track potential risk decision unit is used to set the number of days as the monitoring period P1. When the number of days reaches P1, the potential risk of track management will be subject to decision-making evaluation and analysis. A track potential high-risk signal or a track potential low-risk signal will be generated through analysis, and the track potential high-risk signal or the track potential low-risk signal will be sent to the remote monitoring center. When the remote monitoring center receives the track potential high-risk signal, an alarm will be issued.

[0005] Furthermore, the high performance computing unit includes an image preprocessing module, a target detection module and a foreign body feature extraction and classification module.

[0006] Furthermore, the image preprocessing module performs preprocessing operations including denoising and contrast enhancement on the track image collected by the high-definition camera array unit, and sends the preprocessed track image to the target detection module; the target detection module uses the target detection model trained by the deep learning framework to identify various foreign objects on the track, and the detected and identified target information is sent to the foreign object feature extraction and classification module; The foreign body feature extraction and classification module extracts features from the detected targets. The extracted features include the shape, size and color of the targets. The targets are classified in combination with machine learning algorithms to distinguish between harmless objects and dangerous foreign bodies.

[0007] Furthermore, the specific analysis process of the track potential risk decision-making unit includes: The number of dangerous foreign objects detected and identified during the monitoring period is collected and marked as the total inspection value of dangerous foreign objects, and the monitoring period is divided into several analysis periods, the number of dangerous foreign objects detected and identified during the corresponding analysis period is collected and marked as the period foreign object coefficient, and the period foreign object coefficient is numerically compared with the preset period foreign object coefficient threshold. If the period foreign object coefficient exceeds the preset period foreign object coefficient threshold, the corresponding analysis period is marked as a foreign object intensive period; if the period foreign object coefficient does not exceed the preset period foreign object coefficient threshold, the corresponding analysis period is marked as a foreign object low risk period; The number of foreign body intensive periods in the monitoring period is obtained and marked as the foreign body intensive detection value, and the number of foreign body intensive periods in the interval between two adjacent groups of foreign body low-risk periods is obtained and marked as the intensive continuous value, and the intensive continuous value with the largest value in the monitoring period is marked as the intensive continuous amplitude value; The foreign object impact coefficient is obtained by numerically calculating the total inspection value of dangerous foreign objects, the foreign object density detection value and the density amplitude value, and the foreign object impact coefficient is numerically compared with the preset foreign object impact coefficient threshold. If the foreign object impact coefficient exceeds the preset foreign object impact coefficient threshold, a potential high-risk signal for the track is generated.

[0008] Furthermore, if the foreign object impact coefficient does not exceed the preset foreign object impact coefficient threshold, the emergency processing coefficient and the camera hidden danger coefficient are obtained, and the track potential risk value is obtained by numerically calculating the foreign object impact coefficient, the emergency processing coefficient and the camera hidden danger coefficient; The track potential risk value is numerically compared with the preset track potential risk threshold. If the track potential risk value exceeds the preset track potential risk threshold, a track potential high risk signal is generated; if the track potential risk value does not exceed the preset track potential risk threshold, a track potential low risk signal is generated.

[0009] Furthermore, the track potential risk decision unit is communicatively connected to the emergency processing tracing unit, which analyzes the timeliness of handling track foreign matter during the monitoring period, obtains an emergency processing coefficient through analysis, and sends the emergency processing coefficient to the track potential risk decision unit.

[0010] Furthermore, the specific analysis process of the emergency handling traceability unit is as follows: When the high-performance computing unit detects a dangerous foreign object, the moment when the early warning mechanism is triggered is marked as the danger recognition moment, and the moment when the track maintenance personnel complete the processing of the corresponding dangerous foreign object is marked as the danger removal moment, and the time difference between the danger removal moment and the corresponding danger recognition moment is calculated to obtain the clearance interval length; All the clearing intervals during the monitoring period are obtained and their average is calculated to obtain the clearing efficiency value, and the clearing interval is numerically compared with the preset clearing interval threshold. If the clearing interval exceeds the preset clearing interval threshold, the corresponding clearing interval is marked as the clearing abnormality duration. The number of clearing abnormality durations during the monitoring period is obtained and the ratio is calculated with the number of clearing intervals to obtain the clearing abnormality value. The emergency processing coefficient is obtained by numerically calculating the clearing efficiency value and the clearing abnormality value.

[0011] Furthermore, the track potential risk decision unit is communicatively connected to the camera operation management unit, and the camera operation management unit conducts a comprehensive assessment and analysis of the operational hazards of all high-definition cameras involved in the high-definition camera array unit, obtains a camera hazard coefficient through analysis, and sends the camera hazard coefficient to the track potential risk decision unit.

[0012] Furthermore, the specific analysis process of the camera operation management unit includes: By analyzing the camera operation, a camera risk value of the corresponding high-definition camera is obtained, and the camera risk value is numerically compared with a preset camera risk threshold. If the camera risk value exceeds the preset camera risk threshold, the corresponding high-definition camera is marked as a decaying camera; The number of decaying cameras is obtained and its ratio is calculated with the total number of high-definition cameras to obtain the camera inferiority ratio value, and the camera hazard value of all high-definition cameras is averaged to obtain the camera hazard table value, and the camera hazard value with the largest numerical value is marked as the camera hazard amplitude; the camera hidden danger coefficient is obtained by numerically calculating the camera inferiority ratio value, the camera hazard table value and the camera hazard amplitude.

[0013] Furthermore, the specific analysis process of camera operation analysis is as follows: Obtain the detection data of various environmental parameters that need to be monitored in the environment where the corresponding high-definition camera is located, and compare the detection data of various environmental parameters with the corresponding preset data requirements. If there are environmental parameters that do not meet the corresponding preset data requirements, it is determined that the corresponding camera is in a harmful operating state; Obtain the total duration of the corresponding high-definition camera in the harmful operation state in the historical stage and mark it as the harmful duration, and mark the interval between the production date of the corresponding high-definition camera and the current date as the production duration; The time when the corresponding high-definition camera collects the corresponding track image is obtained, and the time difference between the collection time of the corresponding track image and the reception time of the high-performance computing unit is calculated to obtain the camera delay time, and the average of all camera delay times of the corresponding high-definition camera during the monitoring period is calculated to obtain the camera delay value; The number of times the corresponding high-definition camera is unable to perform video monitoring due to malfunction during the monitoring period is obtained and marked as the camera frequency deviation value, and the camera hazard value of the corresponding high-definition camera is obtained by numerically calculating the harmful time, production time, camera delay value and camera frequency deviation value.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. In the present invention, by introducing machine vision technology, intelligent and automated monitoring of foreign objects on the track is realized, which significantly improves the efficiency and accuracy of railway safety monitoring, effectively reduces the burden of manual inspection, and conducts decision-making evaluation and analysis on the potential risks of track management through the track potential risk decision-making unit. When a potential high-risk signal of the track is generated, the track management personnel are reminded to strengthen the subsequent supervision of the track, so as to further ensure the safety of track operation; 2. In the present invention, the timeliness of handling foreign objects on the track during the monitoring period is analyzed through the emergency handling tracing unit, and the camera operation management unit conducts a comprehensive evaluation and analysis of the operational risks of all high-definition cameras involved in the high-definition camera array unit. It can accurately feedback the handling performance of dangerous foreign objects on the track during the monitoring period and the operational management difficulty of all high-definition cameras, and provide data support for the analysis process of the track potential risk decision-making unit to ensure the comprehensiveness of the analysis and the accuracy of the analysis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings; Figure 1 This is a system block diagram of Embodiment 1 of the present invention; Figure 2 This is a system block diagram of Embodiment 2 and Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] Embodiment 1: Figure 1 As shown, a track foreign body detection system based on machine vision proposed by the present invention includes a high-definition camera array unit, a high-performance computing unit, a track potential risk decision unit and a remote monitoring center; The HD camera array unit consists of several groups of HD cameras installed on both sides or above the track. It adopts high resolution and wide viewing angle design to ensure coverage of the entire track area. It also has night infrared imaging capabilities to adapt to all-weather working environments. The HD camera array unit monitors the track in all directions and sends the collected track images to the high-performance computing unit in real time. The high-performance computing unit processes and analyzes the received track images (responsible for image processing and data analysis, with a built-in high-performance GPU accelerator card to improve the efficiency of algorithm operation), and judges whether there are dangerous foreign objects on the track based on this. When dangerous foreign objects are detected, the early warning mechanism is triggered to generate corresponding early warning information, and the early warning information and the location of the dangerous foreign objects are sent to the remote monitoring center. When the remote monitoring center receives the early warning information, it arranges for track maintenance personnel to go and deal with the dangerous foreign objects.

[0018] Through the introduction of machine vision technology, intelligent and automated monitoring of foreign objects on tracks has been realized, significantly improving the efficiency and accuracy of railway safety monitoring. It can not only timely detect and warn of potential safety hazards, but also effectively reduce the burden of manual inspections, providing strong technical support for the safe operation of railway transportation.

[0019] It should be noted that the high-performance computing unit includes an image preprocessing module, a target detection module, and a foreign body feature extraction and classification module; wherein the image preprocessing module performs preprocessing operations such as denoising and contrast enhancement on the track image collected by the high-definition camera array unit to improve the image quality, and sends the preprocessed track image to the target detection module; The target detection module uses a target detection model trained by a deep learning framework (such as YOLO, Faster R-CNN, etc.) to identify various foreign objects on the track, and the detected and identified target information is sent to the foreign object feature extraction and classification module; The foreign body feature extraction and classification module extracts features from the detected targets. The extracted features include key features such as the target's shape, size, and color. It combines machine learning algorithms (such as SVM, random forest, etc.) to classify the targets and distinguish between harmless objects and dangerous foreign bodies.

[0020] The track potential risk decision unit is used to set the number of days as the monitoring period P1, preferably, P1 = 20; when the number of days reaches P1, the potential risk of track management is analyzed for decision-making assessment, and a track potential high risk signal or a track potential low risk signal is generated through analysis, and the track potential high risk signal or the track potential low risk signal is sent to the remote monitoring center; When the remote monitoring center receives a potential high-risk signal from the track, it will issue an alarm, conduct a comprehensive analysis and comprehensive evaluation of the track management risk, and promptly remind the track management personnel to strengthen the subsequent supervision of the track, so as to ensure the safety of track operation; the specific analysis process of the track potential risk decision unit is as follows: The monitoring period is divided into several analysis periods, each of which has the same duration; the number of dangerous foreign objects detected and identified in the corresponding analysis period is collected and marked as the period foreign object coefficient, and the period foreign object coefficient is numerically compared with the preset period foreign object coefficient threshold. If the period foreign object coefficient exceeds the preset period foreign object coefficient threshold, the corresponding analysis period is marked as a foreign object intensive period; if the period foreign object coefficient does not exceed the preset period foreign object coefficient threshold, the corresponding analysis period is marked as a foreign object low risk period; The number of foreign body intensive periods during the monitoring period is obtained and marked as the foreign body intensive detection value, the number of foreign body intensive periods within the interval between two adjacent groups of foreign body low-risk periods is obtained and marked as the intensive continuous value, the intensive continuous value with the largest value during the monitoring period is marked as the intensive sustained amplitude value, and the number of dangerous foreign bodies detected and identified during the monitoring period is collected and marked as the total dangerous foreign body inspection value; The foreign body influence coefficient HX is obtained by numerically calculating the total inspection value WR of dangerous foreign bodies, the foreign body intensive detection value WX and the intensive amplitude value NP through the formula HX=c1×WR+c2×WX+c3×NP; wherein c1, c2 and c3 are preset weight coefficients with values ​​greater than zero, and the larger the value of the foreign body influence coefficient HX is, the greater the potential management risk of the track is. The foreign body impact coefficient HX is numerically compared with the preset foreign body impact coefficient threshold. If the foreign body impact coefficient HX exceeds the preset foreign body impact coefficient threshold, it indicates that the potential management risk of the track is initially large, and a potential high-risk signal for the track is generated.

[0021] Furthermore, if the foreign body influence coefficient HX does not exceed the preset foreign body influence coefficient threshold, the emergency processing coefficient QF and the camera hidden danger coefficient QS are obtained; The foreign body influence coefficient HX, the emergency treatment coefficient QF and the camera hidden danger coefficient QS are numerically calculated by the formula QX=(su1×HX+su2×QF+su3×QS) / 3 to obtain the track potential risk value QX; among them, su1, su2, su3 are preset weight coefficients with values ​​greater than zero, and the larger the value of the track potential risk value QX, the greater the potential management risk of the track in general; The track potential risk value is numerically compared with the preset track potential risk threshold. If the track potential risk value exceeds the preset track potential risk threshold, it indicates that the potential management risks of the track are generally large and track supervision needs to be strengthened, then a track potential high risk signal is generated; if the track potential risk value does not exceed the preset track potential risk threshold, it indicates that the potential management risks of the track are generally small and then a track potential low risk signal is generated.

[0022] Embodiment 2: Figure 2 As shown, the difference between this embodiment and the first embodiment is that the track potential risk decision unit is connected to the emergency treatment tracing unit in communication. The emergency treatment tracing unit analyzes the timeliness of the treatment of track foreign objects during the monitoring period, obtains the emergency treatment coefficient QF through analysis, and sends the emergency treatment coefficient QF to the track potential risk decision unit, which can not only accurately feedback the treatment performance of track dangerous foreign objects during the monitoring period, but also provide data support for the analysis process of the track potential risk decision unit, ensuring the comprehensiveness of its analysis and the accuracy of the analysis results; the specific analysis process of the emergency treatment tracing unit is as follows: When the high-performance computing unit detects a dangerous foreign object, the moment when the early warning mechanism is triggered is marked as the danger recognition moment, and the moment when the track maintenance personnel complete the processing of the corresponding dangerous foreign object is marked as the danger removal moment, and the time difference between the danger removal moment and the corresponding danger recognition moment is calculated to obtain the clearance interval length; it should be noted that the larger the value of the clearance interval length, the less timely the clearance of the corresponding dangerous foreign object is; All the cleaning intervals in the monitoring period are obtained and their average is calculated to obtain the cleaning efficiency value, and the cleaning interval is compared with the preset cleaning interval threshold. If the cleaning interval exceeds the preset cleaning interval threshold, it indicates that the corresponding dangerous foreign matter is not cleared in time, and the corresponding cleaning interval is marked as the cleaning abnormality time. The number of cleaning abnormality times in the monitoring period is obtained and the ratio is calculated with the number of cleaning intervals to obtain the cleaning abnormality value. The emergency treatment coefficient QF is obtained by numerically calculating the removal efficiency value YP and the removal abnormality value GW through the formula QF=a1×YP+a2×GW; wherein a1 and a2 are preset weight coefficients whose values ​​are greater than zero, and the larger the value of the emergency treatment coefficient QF, the worse the overall handling performance of dangerous foreign objects on the track during the monitoring period, and the greater the adverse impact on the safe operation of the track.

[0023] Embodiment 3: Figure 2 As shown, the difference between this embodiment and the first and second embodiments is that the track potential risk decision unit is connected to the camera operation management unit in communication, and the camera operation management unit conducts a comprehensive evaluation and analysis of the operation hazards of all high-definition cameras involved in the high-definition camera array unit, and obtains the camera hazard coefficient QS through analysis; The camera hidden danger coefficient QS is sent to the track potential risk decision unit, which can not only accurately feedback the monitoring hidden danger degree of all high-definition cameras, but also provide data support for the analysis process of the track potential risk decision unit, further ensuring the comprehensiveness of its analysis and the accuracy of the analysis results; the specific analysis process of the camera operation management unit is as follows: Obtain the detection data of various environmental parameters (including temperature, humidity and other parameters) that need to be monitored in the environment where the corresponding high-definition camera is located, and compare the detection data of various environmental parameters with the corresponding preset data requirements. If there are environmental parameters that do not meet the corresponding preset data requirements, it indicates that the current environment is likely to cause damage to the corresponding high-definition camera, and the corresponding camera is judged to be in a harmful operating state; The total duration of the corresponding high-definition camera in the historical stage being in a harmful operating state is obtained and marked as the harmful duration, and the production date of the corresponding high-definition camera is obtained, and the interval between the production date of the corresponding high-definition camera and the current date is marked as the production duration; The time when the corresponding high-definition camera collects the corresponding track image is obtained, and the time difference between the collection time of the corresponding track image and the reception time of the high-performance computing unit is calculated to obtain the camera delay time, and the average of all the camera delay times of the corresponding high-definition camera during the monitoring period is calculated to obtain the camera delay value; and the number of occurrences in which the corresponding high-definition camera cannot perform camera monitoring due to failure during the monitoring period is obtained and marked as the camera frequency difference value; The harmful time SF, production time SP, camera delay value SM and camera frequency difference value SX are numerically calculated by the formula SY=(e1×SF+e2×SP) / 2+e3×SM+e4×SX to obtain the camera hazard value SY of the corresponding high-definition camera; wherein, e1, e2, e3 and e4 are preset weight coefficients with values ​​greater than zero, and the larger the value of the camera hazard value SY is, the greater the difficulty of operation and management of the corresponding high-definition camera is, and the more difficult it is to ensure the efficient and stable operation of the corresponding high-definition camera; Compare the camera danger value with the preset camera danger threshold. If the camera danger value exceeds the preset camera danger threshold, it indicates that the operation and management of the corresponding high-definition camera is difficult and it is difficult to ensure the efficient and stable operation of the corresponding high-definition camera. Then, the corresponding high-definition camera is marked as a decaying camera. The number of decay cameras is obtained and the ratio thereof is calculated with the total number of high-definition cameras to obtain a camera inferiority value, and the camera risk condition values ​​of all high-definition cameras are averaged to obtain a camera risk table value, and the camera risk condition value with the largest value is marked as a camera risk amplitude; The camera hazard coefficient QS is obtained by numerically calculating the camera occupancy value TM, the camera hazard table value TR and the camera hazard amplitude XP through the formula QS=uy1×TM+(uy2×TR+uy3×XP) / 2; wherein, uy1, uy2, uy3 are preset weight coefficients, uy1>uy2>uy3>0; and, the larger the value of the camera hazard coefficient QS, the greater the overall difficulty of operation and management of all high-definition cameras, and the greater the adverse impact on the safe operation of the track.

[0024] The working principle of the present invention is as follows: when in use, the track is monitored in all directions through a high-definition camera array unit, the high-performance computing unit processes and analyzes the track image and determines whether there are dangerous foreign objects on the track, and when dangerous foreign objects are detected, the early warning mechanism is triggered to generate corresponding early warning information, and when the remote monitoring center receives the early warning information, the track maintenance personnel are arranged to go and handle the dangerous foreign objects. Through the introduction of machine vision technology, intelligent and automated monitoring of track foreign objects is realized, which significantly improves the efficiency and accuracy of railway safety monitoring, effectively reduces the burden of manual inspections, and through the track potential risk decision unit, the potential risk of track management is evaluated and analyzed, and when a potential high-risk signal of the track is generated, the track management personnel are reminded to strengthen the subsequent supervision of the track, so as to further ensure the safety of track operation.

[0025] The above formulas are all dimensionless and numerical calculations. The formula is a formula obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formula are set by technicians in this field according to actual conditions. The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the contents of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that technicians in the relevant technical field can understand and use the present invention well. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A track foreign body detection system based on machine vision, characterized in that: It includes a high-definition camera array unit, a high-performance computing unit, a track potential risk decision unit and a remote monitoring center; the high-definition camera array unit consists of several groups of high-definition cameras, which are installed on both sides or above the track. It adopts a high-resolution and wide-viewing angle design to ensure coverage of the entire track area. It also has night infrared imaging capabilities. The track is fully monitored through the high-definition camera array unit, and the collected track images are sent to the high-performance computing unit in real time; The high-performance computing unit processes and analyzes the received track images to determine whether there are dangerous foreign objects on the track. When dangerous foreign objects are detected, the early warning mechanism is triggered to generate corresponding early warning information, and the early warning information and the location of the dangerous foreign objects are sent to the remote monitoring center. When the remote monitoring center receives the early warning information, it arranges track maintenance personnel to go and handle the dangerous foreign objects. The track potential risk decision unit is used to set the number of days as the monitoring period P1. When the number of days reaches P1, the potential risk of track management will be subject to decision-making evaluation and analysis. A track potential high-risk signal or a track potential low-risk signal will be generated through analysis, and the track potential high-risk signal or the track potential low-risk signal will be sent to the remote monitoring center. When the remote monitoring center receives the track potential high-risk signal, an alarm will be issued.

2. According to the machine vision-based track foreign body detection system of claim 1, it is characterized in that: The high-performance computing unit includes an image preprocessing module, a target detection module and a foreign body feature extraction and classification module.

3. A track foreign body detection system based on machine vision according to claim 2, characterized in that: The image preprocessing module performs preprocessing operations including denoising and contrast enhancement on the track image collected by the high-definition camera array unit, and sends the preprocessed track image to the target detection module; The target detection module uses the target detection model trained by the deep learning framework to identify various foreign objects on the track, and the detected and identified target information is sent to the foreign object feature extraction and classification module; The foreign body feature extraction and classification module extracts features from the detected targets. The extracted features include the shape, size and color of the targets. The targets are classified in combination with machine learning algorithms to distinguish between harmless objects and dangerous foreign bodies.

4. The track foreign body detection system based on machine vision according to claim 1, characterized in that: The specific analysis process of the track potential risk decision-making unit includes: The number of dangerous foreign objects detected and identified during the monitoring period is collected and marked as the total inspection value of dangerous foreign objects, and the monitoring period is divided into several analysis periods, and the number of dangerous foreign objects detected and identified during the corresponding analysis period is collected and marked as the period foreign object coefficient. If the period foreign object coefficient exceeds the preset period foreign object coefficient threshold, the corresponding analysis period is marked as a foreign object intensive period; if the period foreign object coefficient does not exceed the preset period foreign object coefficient threshold, the corresponding analysis period is marked as a foreign object low risk period; The number of foreign body intensive periods in the monitoring period is obtained and marked as the foreign body intensive detection value, and the number of foreign body intensive periods in the interval between two adjacent groups of foreign body low-risk periods is obtained and marked as the intensive continuous value, and the intensive continuous value with the largest value in the monitoring period is marked as the intensive continuous amplitude value; The foreign object impact coefficient is obtained by numerically calculating the total inspection value of dangerous foreign objects, the foreign object intensive detection value and the intensive amplitude value. If the foreign object impact coefficient exceeds the preset foreign object impact coefficient threshold, a potential high-risk signal for the track is generated.

5. The track foreign body detection system based on machine vision according to claim 4, characterized in that: If the foreign body impact coefficient does not exceed the preset foreign body impact coefficient threshold, the emergency handling coefficient and the camera hidden danger coefficient are obtained, and the track potential risk value is obtained by numerically calculating the foreign body impact coefficient, the emergency handling coefficient and the camera hidden danger coefficient; If the track potential risk value exceeds the preset track potential risk threshold, a track potential high risk signal is generated; If the track potential risk value does not exceed the preset track potential risk threshold, a track potential low risk signal is generated.

6. The track foreign body detection system based on machine vision according to claim 4, characterized in that: The track potential risk decision unit is communicated with the emergency handling tracing unit. The emergency handling tracing unit analyzes the timeliness of handling track foreign matter during the monitoring period, obtains the emergency handling coefficient through analysis, and sends the emergency handling coefficient to the track potential risk decision unit.

7. The track foreign body detection system based on machine vision according to claim 6, characterized in that: The specific analysis process of the emergency handling traceability unit is as follows: When the high-performance computing unit detects a dangerous foreign object, the moment when the early warning mechanism is triggered is marked as the danger recognition moment, and the moment when the track maintenance personnel complete the processing of the corresponding dangerous foreign object is marked as the danger removal moment, and the time difference between the danger removal moment and the corresponding danger recognition moment is calculated to obtain the clearance interval length; All the clearing intervals during the monitoring period are obtained and their average is calculated to obtain the clearing efficiency value, and the clearing interval is numerically compared with the preset clearing interval threshold. If the clearing interval exceeds the preset clearing interval threshold, the corresponding clearing interval is marked as the clearing abnormality duration. The number of clearing abnormality durations during the monitoring period is obtained and the ratio is calculated with the number of clearing intervals to obtain the clearing abnormality value. The emergency processing coefficient is obtained by numerically calculating the clearing efficiency value and the clearing abnormality value.

8. The track foreign body detection system based on machine vision according to claim 6, characterized in that: The track potential risk decision unit is communicatively connected to the camera operation management unit. The camera operation management unit conducts a comprehensive assessment and analysis of the operational hazards of all high-definition cameras involved in the high-definition camera array unit, obtains a camera hazard coefficient through analysis, and sends the camera hazard coefficient to the track potential risk decision unit.

9. The track foreign body detection system based on machine vision according to claim 8, characterized in that: The specific analysis process of the camera operation management unit includes: By running the camera analysis, a camera danger value of the corresponding high-definition camera is obtained. If the camera danger value exceeds a preset camera danger threshold, the corresponding high-definition camera is marked as a decaying camera. The number of decaying cameras is obtained and its ratio is calculated with the total number of high-definition cameras to obtain the camera inferiority ratio value, and the camera hazard value of all high-definition cameras is averaged to obtain the camera hazard table value, and the camera hazard value with the largest numerical value is marked as the camera hazard amplitude; the camera hidden danger coefficient is obtained by numerically calculating the camera inferiority ratio value, the camera hazard table value and the camera hazard amplitude.

10. A track foreign body detection system based on machine vision according to claim 9, characterized in that: The specific analysis process of camera operation analysis is as follows: Obtain the detection data of various environmental parameters that need to be monitored in the environment where the corresponding high-definition camera is located. If there are environmental parameters that do not meet the corresponding preset data requirements, it is determined that the corresponding camera is in a harmful operating state; Obtain the total duration of the corresponding high-definition camera in the harmful operation state in the historical stage and mark it as the harmful duration, and mark the interval between the production date of the corresponding high-definition camera and the current date as the production duration; The time when the corresponding high-definition camera collects the corresponding track image is obtained, and the time difference between the collection time of the corresponding track image and the reception time of the high-performance computing unit is calculated to obtain the camera delay time, and the average of all camera delay times of the corresponding high-definition camera during the monitoring period is calculated to obtain the camera delay value; The number of times the corresponding high-definition camera is unable to perform video monitoring due to malfunction during the monitoring period is obtained and marked as the camera frequency deviation value, and the camera hazard value of the corresponding high-definition camera is obtained by numerically calculating the harmful time, production time, camera delay value and camera frequency deviation value.

Images