Precise sleeper counting and positioning correction method for subway inspection system

Abstract

The invention discloses a sleeper accurate counting and positioning correction method for a subway inspection system, and relates to the technical field of rail traffic intelligent inspection. Detecting sleepers through a first feature extraction network and counting in real time, and filtering incomplete sleepers with the area less than 1 / 2 of the complete sleepers to avoid repeated counting; a mileage number spraying area is positioned through a second feature extraction network, character recognition is triggered, absolute position information contained in number spraying is analyzed, and the current sleeper counting value is corrected through the absolute position information and then continues to be accumulated; meanwhile, inter-station information is obtained through real-time communication to serve as interval constraints, and positioning errors are restrained in a non-spraying area; when the number spraying is not recognized in the continuous preset distance, the counting result is corrected again; and the output position signal can be directly used for guiding maintenance equipment or personnel to quickly arrive at a defect point, so that the manual recheck time is shortened, the repeated inspection and storage occupation are reduced, and the train operation safety is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection technology for rail transit, and in particular to a method for accurate counting and positioning correction of sleepers for subway inspection systems. Background Technology

[0002] As a vital component of urban public transportation, the safe and stable operation of subways directly impacts citizen safety and urban traffic efficiency. With the continuous increase in line scale and passenger volume, subway systems are exhibiting high-density, high-load operation characteristics, placing higher demands on the timeliness of line inspections and the accuracy of hazard handling. Precise location of hazards is fundamental for subsequent handling and repair: under conditions of short operating intervals and high downtime costs, once a hazard is discovered, reliable positioning methods are needed to quickly pinpoint its exact location, buying time for repairs and dispatching, and minimizing the impact on normal operations.

[0003] Current subway line inspection and positioning methods mainly include manual marking, satellite positioning, and wheel axle pulse counting. These methods generally have limitations: manual marking is inefficient, has limited inspection mileage coverage, and is easily affected by human experience and recording errors; satellite positioning suffers from unstable signals in tunnels, platforms, and obstructed environments, resulting in significant positioning errors and failing to meet the precise positioning requirements of subway scenarios; inertial navigation units typically rely on accelerometers and gyroscopes to calculate displacement integration, making them susceptible to zero-bias drift, noise accumulation, installation errors, and train vibrations, causing positioning errors to drift continuously over time and mileage, and making long-term stable correction difficult in the absence of external absolute references; wheel axle pulse counting is easily affected by wheel slippage, rail wear, and wheel diameter changes, with errors increasing with mileage accumulation, leading to significant positioning deviations after long-term operation. Some positioning solutions use RFID tags for mileage calibration or location verification. However, RFID requires the deployment of a large number of tags along the route and the matching reading and writing equipment, resulting in high construction and maintenance costs. Tags are susceptible to moisture, dust, metal structures, and electromagnetic environments, leading to fluctuations in reading and writing distances or missed or misreadings. Furthermore, RFID tags are usually discretely distributed, and distance calculations are still needed between points, making it difficult to achieve continuous high-precision positioning. Due to insufficient positioning accuracy, potential hazards are often difficult to locate in a short time, thus extending maintenance windows, increasing operating costs, and even posing potential risks to driving safety. Summary of the Invention

[0004] The purpose of this invention is to provide a method for accurate counting and positioning correction of sleepers in a subway inspection system, which solves the problems of cumulative errors caused by relative positioning that relies solely on sleeper counting; insufficient stability of mileage marking identification, making it difficult to form reliable absolute anchor points; and uncertainty in segment attribution and limited onboard computing power when absolute anchor points are missing near stations or in areas without markings.

[0005] To achieve the above objectives, the present invention provides a method for accurate counting and positioning correction of sleepers in a subway inspection system, comprising the following steps: Step 1: The onboard industrial control computer on the data acquisition platform establishes a connection with the Metro TCMS through the internal communication interface to obtain the line number, driving direction, current station ID, and next station ID; Step 2: Start the vehicle-mounted linear array imaging system. The linear array camera, triggered by the driven wheel encoder, adaptively acquires images of the track surface strips according to the vehicle speed. The continuously acquired strip images are stitched together in time sequence to reconstruct a complete two-dimensional detection image of the track surface, and transmitted to the vehicle-mounted industrial control computer in real time. Step 3: The onboard industrial control computer performs sleeper detection and counting on each two-dimensional detection image of the track surface; Step 4: The onboard industrial control computer performs inkjet marking recognition on each two-dimensional detection image of the track surface. Step 5: Based on the identified inkjet marking target, the global sleeper count is corrected for consistency.

[0006] Furthermore, the vehicle-mounted linear array imaging system specifically utilizes the forward direction of the track image acquisition platform to scan and image the track. The track image acquisition platform consists of three 2D line scan cameras and three line laser light sources. The projection plane of the line laser light source is projected perpendicularly onto the track bed. The camera imaging plane scans and images the track strips illuminated by the line laser. The continuously acquired strip images are stitched together in a time sequence to reconstruct a complete two-dimensional detection image of the track surface.

[0007] Furthermore, the method for the on-board industrial control computer to perform sleeper detection on each two-dimensional detection image of the track surface is as follows: input the two-dimensional detection image of the track surface into the first feature extraction network to obtain sleeper candidate targets.

[0008] Furthermore, the method for the onboard industrial control computer to count sleepers for each two-dimensional detection image of the track surface employs an integrity filtering method. Specifically, it calculates the effective area for each target. and the reference area of ​​the entire track Comparison: If If it is determined to be an incomplete sleeper, it will be discarded; if It is then treated as a countable sleeper, and the global sleeper count is incremented by 1 when the condition for its first appearance is met.

[0009] Furthermore, the method for the on-board industrial control computer to perform inkjet number recognition on each two-dimensional detection image is as follows: input the two-dimensional detection image of the track surface into the second feature extraction network to locate the inkjet number area, crop it according to the detection box and input it into the OCR module to obtain the inkjet number string, and parse the inkjet number string to obtain the absolute position information of the driving direction, kilometer marker and sleeper number.

[0010] Furthermore, both the first and second feature extraction networks are lightweight dynamic detection models, employing an improved YOLOv11 architecture. The improvement lies in replacing the C3K2 module used for feature extraction in the YOLOv11 backbone network with a heavily parameterized lightweight backbone network, RepGhostNet; and introducing the TrackLargeObjMCA module in the neck or feature fusion stage.

[0011] Furthermore, the OCR module incorporates a specific OCR algorithm for optical character recognition.

[0012] Furthermore, the method for consistency correction of the global sleeper count based on the identified spray mark targets includes: Step 501: Record the sleeper count value corresponding to the obtained sleeper serial number as... The current cumulative global sleeper count is recorded as And directly at the moment the spray mark character is detected. right Perform correction and update; Step 502: After the update is complete, use the corrected version. As a new counting benchmark, the subsequent complete sleepers will continue to be incrementally accumulated; Step 503: Write the current station ID and the next station ID as interval constraints into the positioning record, and output "Traffic direction - Inter-station interval - Kilometer marker - Nth sleeper".

[0013] Furthermore, it also includes: Step 601: The on-board industrial control computer communicates with the TCMS in real time to obtain at least the following information: route number, driving direction, current station ID, next station ID, and arrival and departure status signals; Step 602: When the outbound status signal is received, set the current section to ( ), Indicates the current station number. Indicates the next station number, and reads the constraint parameters for that interval from the preset inter-station database. ), This indicates the number of sleepers corresponding to the current station. This indicates the sleeper value corresponding to the next station, and initializes the sleeper count for the section to [value]. And count the global sleepers Establish a correspondence between the track sleeper count and the track section count; Step 603: In the no-spray-number constraint mode, perform integrity filtering on the sleeper detection results of each image first. Only sleepers that meet the conditions are allowed to trigger count increment. The global sleeper count is based on... Output; Step 604: When the TCMS reports the arrival status signal, update ( ), ( And initialize the sleeper count for the section to Continue processing subsequent images Continuously count and output the global number of sleepers. ; Step 605: When the spray mark is re-identified in a subsequent frame, parse the absolute position information contained in the spray mark and obtain its corresponding sleeper serial number. Use spray number Current global sleeper count Perform corrections.

[0014] Furthermore, the system is considered to enter the no-spray-number constraint mode if any of the following conditions are met: No spray mark characters were detected within several consecutive frames / time windows; Track sleeper count Exceeding the preset threshold between stations ; The continuous driving distance exceeds the preset distance threshold.

[0015] Therefore, the present invention employs the above-mentioned method for accurate counting and positioning correction of sleepers for subway inspection systems, which has the following beneficial effects: 1. By correcting the consistency between the absolute anchor point determined by the inkjet number analysis and the sleeper count, a positioning correction closed loop is formed, which significantly suppresses the cumulative error caused by missed detections and frame loss; 2. By filtering out incomplete sleepers with an area of ​​less than 1 / 2, the probability of duplicate counting is reduced, thus improving the stability of sleeper counting; 3. Combine TCMS inter-station information to perform interval constraints, and use the station database and arrival information for compensation and correction to avoid location loss near the station or in areas without inkjet numbers; 4. Outputting location information can directly guide maintenance equipment or personnel to quickly reach the defect point, reducing manual review time and repeated troubleshooting; 5. By utilizing vehicle-mounted linear array vision and TCMS information, there is no need to deploy additional RFID, high-precision inertial navigation, GNSS, etc. along the route, reducing deployment and maintenance costs and reducing the storage of redundant invalid images.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for accurate counting and positioning correction of sleepers for a subway inspection system according to the present invention; Figure 2 This is a diagram showing the arrangement of the 2D line scan camera and line laser source of the present invention. Figure 3 This is a diagram showing the camera acquisition effect of the present invention; Figure 4 This is a structural diagram of the lightweight dynamic detection model of the present invention; Figure 5 This is a diagram showing the detection effect of the lightweight dynamic detection model of the present invention; Figure 6 This is a flowchart of the text recognition process of the present invention; Figure 7 This is a flowchart of the mileage correction process of the present invention; Figure 8 This is a diagram showing the results of defect localization in this invention; Figure 9 This is a comparison chart showing the presence and absence of mileage marking correction in this invention. Detailed Implementation

[0018] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely illustrates selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0019] Please see Figure 1-9 A method for accurate counting and positioning correction of sleepers in a subway inspection system includes the following steps: Step 1: The onboard industrial control computer on the data acquisition platform establishes a connection with the Metro TCMS through the internal communication interface to obtain the line number, driving direction, current station ID, and next station ID; In embodiments of the present invention, the driving direction includes uphill and downhill.

[0020] The Metro TCMS (Train Control and Management System) is the core integrated control system of metro trains, responsible for centralized control, real-time monitoring, fault diagnosis and information management of the entire train.

[0021] Step 2: Start the vehicle-mounted linear array imaging system. The linear array camera, triggered by the driven wheel encoder, adaptively acquires images of the track surface strips according to the vehicle speed. The continuously acquired strip images are stitched together in time sequence to reconstruct a complete two-dimensional detection image of the track surface, and transmitted to the vehicle-mounted industrial control computer in real time. It should be noted that the vehicle-mounted linear array imaging system specifically utilizes the forward direction of the track image acquisition platform to scan and image the track. A schematic diagram of the track image acquisition platform structure is shown below. Figure 2As shown, the track image acquisition platform consists of three 2D line scan cameras and three line laser sources. The projection plane of the line laser sources projects perpendicularly onto the track bed, and the camera imaging plane scans and images the track strip illuminated by the line lasers. The cameras and line laser sources are 500mm above the ground. The camera resolution is 2048×1, and the line frequency is 52 kHz. The center wavelength of the line laser sources is 915nm, and the line width is 5mm-8mm. The camera outputs a continuous two-dimensional image of the track strip with a resolution of 2048×1024, as shown. Figure 3 As shown.

[0022] The camera triggering device is the encoder on the driven wheel of the subway. The line scan camera automatically adjusts the image acquisition triggering frequency according to the pulse signal of the encoder on the driven wheel of the train (such as triggering once every time it turns an angle). It also adaptively adjusts the camera acquisition frequency with the train speed to ensure that the acquired images are clear and continuous.

[0023] Step 3: The onboard industrial control computer performs sleeper detection and counting on each two-dimensional detection image of the track surface; The two-dimensional detection image of the track surface is input into the first feature extraction network to obtain candidate sleeper targets; the detection result is shown in the figure below. Figure 5 As shown. An integrity filtering method is used for counting, which involves calculating the effective area for each target. and the reference area of ​​the entire track Comparison: If If it is determined to be an incomplete sleeper, it will be discarded; if It is then treated as a countable sleeper, and the global sleeper count is incremented by 1 when the condition for its first appearance is met.

[0024] Step 4: The onboard industrial control computer performs inkjet marking recognition on each two-dimensional detection image of the track surface; The two-dimensional detection image of the track surface is input into the second feature extraction network to locate the spray mark area. After being cropped according to the detection frame, it is input into the OCR module (such as the PPOCR series, and PPOCRV4 is used in this embodiment) to obtain the spray mark string. The spray mark string is parsed to obtain the absolute position information such as the driving direction, kilometer marker and sleeper number.

[0025] like Figure 6 As shown, the track sleeper position spray mark collected in this embodiment is S030001 as an example. It is a standardized spray mark code for subway lines, where: S represents the line number for the up / down line, 03 is the kilometer mark value, and 0001 is the sleeper serial number under the corresponding kilometer mark. This parsing logic can be adapted to the spray mark code rules of different subway lines. Only the corresponding parameters of the parsing module need to be adjusted.

[0026] It should be noted that the method for obtaining absolute position information by parsing the spray code string is as follows: split the spray code string and compare it with the preset parsing database to obtain the absolute position information.

[0027] In embodiments of the present invention, both the first feature extraction network and the second feature extraction network are lightweight dynamic detection models, employing the improved YOLOv11 architecture, with the structure as follows: Figure 4 As shown, the C3K2 module for feature extraction in the backbone network is replaced with the heavily parameterized lightweight backbone network RepGhostNet to reduce the number of parameters and computational cost; the SPPF (Spatial Pyramid Pooling) and C2PSA (Spatial Attention) modules of YOLOv11 are retained to ensure the ability of multi-scale feature extraction and spatial attention enhancement. The TrackLargeObjMCA (Large Target Tracking Multidimensional Collaborative Attention) module is introduced in the neck or feature fusion stage to enhance the directional linear structural features of the sleepers, thereby improving the detection stability of sleepers and sleeper position markings; the rest remains unchanged.

[0028] Step 5: Consistency Correction and Interval Constraints; During train operation, fluctuations in train speed, vibrations, encoder jitter, and interference from trigger signals can cause image acquisition quality abnormalities such as incomplete image acquisition, insufficient number of lines, missing lines, and frame loss. This can lead to intermittent missed detections or counting errors in sleeper inspection. Figure 7 As shown, the following steps are used for consistency correction: Step 501: Record the sleeper count value corresponding to the obtained sleeper serial number as... The current cumulative global sleeper count is recorded as And directly at the moment the spray mark character is detected. right Perform a correction and update.

[0029] Step 502: After the update is complete, use the corrected version. As a new counting benchmark, the count is continuously incremented for subsequent complete sleepers. This ensures that if counting deviations occur due to incomplete data acquisition, frame loss fluctuations, or missed detections, they can be eliminated in time when the serial number appears, guaranteeing that the output positioning results are continuous, stable, and traceable.

[0030] Step 503: Write the current station ID and the next station ID as interval constraints into the positioning record, and output "Traffic direction - Inter-station interval - Kilometer marker - Nth sleeper".

[0031] In certain sections of the line, especially areas immediately after leaving a station, where curves obstruct the view, or where the mileage markings are missing or covered by dirt, the mileage markings may not be reliably identified over a considerable distance. To prevent long-term drift in sleeper counts due to missing markings, this embodiment introduces an inter-station database and combines it with TCMS station information to establish section constraints, achieving stable positioning and subsequent correction even under conditions of missing markings.

[0032] If the spray mark is not detected within a consecutive preset distance (e.g., 1km), the vehicle-mounted industrial control computer performs a secondary correction based on the station database and TCMS arrival events: upon receiving arrival information, the count is aligned to the baseline value at the end of the interval to ensure continuity and traceability across intervals. This specifically includes the following steps: Step 601: The on-board industrial control computer communicates with the TCMS in real time to obtain at least the following information: route number, driving direction, current station ID, next station ID, and arrival and departure status signals.

[0033] It should be noted that the onboard industrial control computer pre-stores the inter-station database. The inter-station database is indexed at least by line number, direction of travel, and station number, and contains the corresponding interval parameters for adjacent station pairs. The interval parameters include: the starting and ending station numbers of the interval, the sleeper numbers corresponding to the stations in the interval, and other information.

[0034] In embodiments of the present invention, the departure status signal includes a "departure / departure" event or a current station update event, and the arrival status signal includes a "arrival / stop" event or an update to the next station.

[0035] Step 602: When the outbound status signal is received, set the current section to ( ), Indicates the current station number. Indicates the next station number, and reads the constraint parameters for that interval from the inter-station database. ), This indicates the number of sleepers corresponding to the current station. This indicates the sleeper count for the next station. Simultaneously, the sleeper count for the section is initialized to... And count the global sleepers Establish a correspondence with the sleeper count of the section (for example, use the "section start point" as the basis for incremental counting, and you can output the result of "inter-station section + Nth sleeper" without calculating the mileage value in the unmarked stage).

[0036] During continuous image processing, the "no inkjet number constraint mode" will be entered if any of the following conditions are met: (1) No spray mark characters were detected within several consecutive frames / time windows; (2) Count of sleepers in the section Exceeding the preset threshold between stations ; (3) The continuous driving distance exceeds the preset distance threshold (e.g., 1km, the corresponding distance can be obtained by the encoder pulse accumulation).

[0037] After entering the "no-spray-number constraint mode", the sleepers are still detected and counted, but at the same time, the upper limit constraint between stations is introduced to suppress the accumulation of positioning errors.

[0038] Step 603: In the no-spray-number constraint mode, integrity filtering is first performed on the sleeper detection results of each image to avoid duplicate edge counting. Only sleepers that meet the integrity condition are allowed to trigger count increment. Global sleeper count is... Output.

[0039] Step 604: Closed-loop alignment and interval switching of arrival events; When the TCMS reports the arrival status signal, interval closed-loop alignment is performed. Update ( ), ( And initialize the sleeper count for the section to Continue processing subsequent images. Continuous counting. Output the global number of sleepers. To ensure the total number of sleepers exist to between.

[0040] Step 605: Restore correction when the inkjet number reappears.

[0041] When the jet number is re-identified in a subsequent frame, the absolute position information contained in the jet number is parsed to obtain its corresponding sleeper number. At this point, instead of relying on historical estimates, the spray code is used directly. Current global sleeper count Perform corrections.

[0042] In an embodiment of the present invention, a method for accurate counting and positioning correction of sleepers for a subway inspection system further includes defect positioning. Combined with the aforementioned processing flow, corresponding processing logic is automatically executed based on the detected target category: if a sleeper target is detected, the target is extracted and counted in real time; if a serial number target is detected, the area is extracted for text recognition to complete mileage consistency correction and establish a high-precision positioning benchmark; if a defect candidate target is detected, the confidence probability of it belonging to various types of defects is obtained, the highest probability value is selected, and it is determined whether it exceeds a preset confidence threshold (the preset threshold is 0.5). If the maximum value exceeds the preset threshold, the specific defect type corresponding to the image is determined; if it does not exceed the threshold, it is determined to be in a normal state and is not recorded. When a real defect target is determined to exist, the corrected positioning record corresponding to the time of the defect occurrence is directly called, and the location information used to guide maintenance equipment or personnel is output and saved. Figure 8 As shown, the saved information includes defect type, timestamp, line direction, station section, kilometer marker, and information for the corresponding Nth sleeper. Defect types include: missing elastic clips, displaced elastic clips, broken elastic clips, missing bolts, missing spiral spikes, missing gauge blocks, rail cracks, rail spalling, rail breakage, foreign objects in the ballast bed, and ballast bed cracks, etc.

[0043] The detection results of 1109 images from approximately 1.2km of line are as follows. Figure 9 The processing time for 1109 images was 7.95 seconds, recording a total of 1978.5 sleepers. The model inference speed was 139.5 fps, and this detection model meets the operating speed requirements of subways from 0-140 km / h. After manual verification, the actual number of sleeper blocks in the 1109 images was 1984, with a sleeper counting accuracy of 99.7%. When correcting for sleeper counts at the 1km mark without sleeper markings, the sleeper count accumulated infinitely and could not be mapped to the line mileage. When correcting for sleeper counts with markings, the sleeper count was corrected at the 1km mark using the sleeper serial number from the markings, and the sleeper count was reset to zero every kilometer without error, and then counted correctly.

[0044] This invention addresses the sleeper positioning needs in subway tunnels without satellite signals by proposing a method for accurate sleeper counting and positioning correction that integrates machine vision detection, OCR recognition, and TCMS inter-station information constraints. Specifically, it constructs a three-level positioning constraint system: visual detection, absolute anchor point correction, and interval constraint compensation. This involves using an incomplete sleeper area filtering mechanism to suppress repeated counting errors and improve counting accuracy; using mileage markings as absolute anchor points to correct accumulated sleeper counting deviations in real time; and introducing TCMS inter-station information constraints and station database compensation to achieve continuous positioning across the entire interval. Simultaneously, a lightweight neural network architecture is employed to ensure high-speed inference under the limited computing power of the onboard industrial control computer, meeting the engineering requirements of high-speed subway operation scenarios. This effectively solves the core problems of error accumulation and insufficient onboard real-time performance in traditional positioning methods.

[0045] Compared with existing technologies, this invention does not require additional positioning hardware such as RFID and inertial navigation. By utilizing the on-board linear array imaging system and existing TCMS information, it significantly reduces deployment and maintenance costs. The structured location information output, which is "line direction - station section - kilometer marker - Nth sleeper", can directly guide maintenance personnel to quickly locate defect points, reduce manual review time and repeated inspection costs, improve subway inspection efficiency and train operation safety, and has strong engineering applicability.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for accurate counting and positioning correction of sleepers in a subway inspection system, characterized in that, Includes the following steps: Step 1: The onboard industrial control computer on the data acquisition platform establishes a connection with the Metro TCMS through the internal communication interface to obtain the line number, driving direction, current station ID, and next station ID; Step 2: Start the vehicle-mounted linear array imaging system. The linear array camera, triggered by the driven wheel encoder, adaptively acquires images of the track surface strips according to the vehicle speed. The continuously acquired strip images are stitched together in time sequence to reconstruct a complete two-dimensional detection image of the track surface, and transmitted to the vehicle-mounted industrial control computer in real time. Step 3: The onboard industrial control computer performs sleeper detection and counting on each two-dimensional detection image of the track surface; Step 4: The onboard industrial control computer performs inkjet marking recognition on each two-dimensional detection image of the track surface; Step 5: Perform consistency correction on the global sleeper count based on the identified spray mark targets.

2. The method for accurate counting and positioning correction of sleepers for a subway inspection system according to claim 1, characterized in that, The vehicle-mounted linear array imaging system specifically uses the forward direction of the track image acquisition platform to scan and image the track. The track image acquisition platform consists of three 2D line scan cameras and three line laser light sources. The projection plane of the line laser light source is projected perpendicularly onto the track bed, and the camera imaging plane scans and images the track strip illuminated by the line laser.

3. The method for accurate counting and positioning correction of sleepers in a subway inspection system according to claim 2, characterized in that, The method for the on-board industrial control computer to perform sleeper detection on each two-dimensional detection image of the track surface is as follows: input the two-dimensional detection image of the track surface into the first feature extraction network to obtain sleeper candidate targets.

4. The method for accurate counting and positioning correction of sleepers in a subway inspection system according to claim 3, characterized in that, The onboard industrial control computer performs sleeper counting on each two-dimensional detection image of the track surface using an integrity filtering method. Specifically, it calculates the effective area for each target. and the reference area of ​​the entire track Comparison: If If it is determined to be an incomplete sleeper, it will be discarded; if It is then treated as a countable sleeper, and the global sleeper count is incremented by 1 when the condition for its first appearance is met.

5. The method for accurate counting and positioning correction of sleepers in a subway inspection system according to claim 4, characterized in that, The method for the vehicle-mounted industrial control computer to perform inkjet marking recognition on each two-dimensional detection image of the track surface is as follows: input the two-dimensional detection image of the track surface into the second feature extraction network to locate the inkjet marking area, crop it according to the detection box and input it into the OCR module to obtain the inkjet marking string, and parse the inkjet marking string to obtain the absolute position information of the driving direction, kilometer marker and sleeper number.

6. The method for accurate counting and positioning correction of sleepers in a subway inspection system according to claim 5, characterized in that, Both the first and second feature extraction networks are lightweight dynamic detection models, using an improved YOLOv11 architecture. The improvement lies in replacing the C3K2 module used for feature extraction in the YOLOv11 backbone network with a heavily parameterized lightweight backbone network, RepGhostNet; and introducing the TrackLargeObjMCA module in the neck or feature fusion stage.

7. The method for accurate counting and positioning correction of sleepers for a subway inspection system according to claim 6, characterized in that, The OCR module has a built-in specific OCR algorithm for optical character recognition.

8. The method for accurate counting and positioning correction of sleepers in a subway inspection system according to claim 7, characterized in that, Methods for consistency correction of global sleeper counts based on identified jet number targets include: Step 501: Record the sleeper count value corresponding to the obtained sleeper serial number as... The current cumulative global sleeper count is recorded as And directly at the moment the spray mark character is detected. right Perform correction and update; Step 502: After the update is complete, use the corrected version. As a new counting benchmark, the subsequent complete sleepers will continue to be incrementally accumulated; Step 503: Write the current station ID and the next station ID as interval constraints into the positioning record, and output "Traffic direction - Inter-station interval - Kilometer marker - Nth sleeper".

9. A method for accurate counting and positioning correction of sleepers for a subway inspection system according to claim 8, characterized in that, Also includes: Step 601: The on-board industrial control computer communicates with the TCMS in real time to obtain at least the following information: route number, driving direction, current station ID, next station ID, and arrival and departure status signals; Step 602: When the outbound status signal is received, set the current section to ( ), Indicates the current station number. Indicates the next station number, and reads the constraint parameters for that interval from the preset inter-station database. ), This indicates the number of sleepers corresponding to the current station. This indicates the sleeper value corresponding to the next station, and initializes the sleeper count for the section to [value]. And count the global sleepers Establish a correspondence between the track sleeper count and the track section count; Step 603: In the no-spray-number constraint mode, perform integrity filtering on the sleeper detection results of each image first. Only sleepers that meet the conditions are allowed to trigger count increment. The global sleeper count is based on... Output; Step 604: When the TCMS reports the arrival status signal, update ( ), ( And initialize the sleeper count for the section to Continue processing subsequent images Continuously count and output the global number of sleepers. ; Step 605: When the spray mark is re-identified in a subsequent frame, parse the absolute position information contained in the spray mark and obtain its corresponding sleeper serial number. Use spray number Current global sleeper count Perform corrections.

10. The method for accurate counting and positioning correction of sleepers in a subway inspection system according to claim 9, characterized in that, The system is considered to enter the no-serial-number constraint mode if any of the following conditions are met: No spray mark characters were detected within several consecutive frames / time windows; Track sleeper count Exceeding the preset threshold between stations ; The continuous driving distance exceeds the preset distance threshold.

Images