Video intelligent evaluation method and system for locomotive and train vehicle preparation and maintenance operation
By using deep learning and computer vision technologies, the preparation and maintenance of locomotives and trains are automatically evaluated, solving the problems of low efficiency and poor accuracy in existing technologies, and achieving efficient and accurate detection of safety hazards and quality assessment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHENGZHOU SHENGDI INFORMATION TECH
- Filing Date
- 2026-02-10
- Publication Date
- 2026-06-09
AI Technical Summary
In the preparation and maintenance operations of railway locomotives and rolling stock depots, existing technologies rely on manual analysis of operation videos, which is inefficient and inaccurate, making it difficult to effectively identify potential safety hazards.
By employing a deep fusion of computer vision analysis and deep learning technologies, image data is collected using handheld tools by individual soldiers. Target detection models and specific algorithms are then used to automatically assess the quality of operations and the status of equipment, generating structured reports.
It improved the efficiency and accuracy of operational video analysis, reduced manual labor intensity, enhanced the ability to detect safety hazards, and enabled automated quantitative assessment of locomotive and train quality.
Smart Images
Figure CN122175430A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent operation and maintenance technology for rail transit, and in particular to a video-based intelligent evaluation method and system for locomotive and train preparation and maintenance operations. It is applicable to standardized and intelligent operations and management in railway locomotive depots, rolling stock depots, track maintenance depots, and more broadly in the rail transit industry for locomotive and train preparation and maintenance. Background Technology
[0002] In the domestic railway locomotive, rolling stock, and track maintenance sectors, many aspects of locomotive and train maintenance currently rely on manual operation. To ensure the quality of work by maintenance personnel and the safe operation of locomotives and trains, handheld tools such as camera flashlights, recorders, and industrial mobile phones have been introduced into these processes. Clear standardized operating procedures have been developed for maintenance positions to ensure that personnel fully utilize these tools. To further promote the implementation of these standardized procedures, prevent omissions, and assist in identifying problems, management assigns personnel to conduct full-process analysis of the videos collected by each worker using these tools. However, in practice, relying on manual analysis for the large volume of videos generated daily is inefficient, and the accuracy of the analysis cannot be consistently guaranteed. How to leverage advanced productivity tools to conveniently assist maintenance and repair personnel in completing the entire operation process with high quality, improve analysis efficiency and accuracy, and help promptly identify potential safety hazards in locomotives and trains is an urgent problem to be solved. Summary of the Invention
[0003] To address the pain points of low efficiency, poor accuracy, and easy omission of safety hazards in analyzing and judging the condition of locomotives and train equipment in scenarios that heavily rely on manual labor for preparation and maintenance operations, this invention provides a video-based intelligent evaluation method and system for locomotive and train preparation and maintenance operations. By deeply integrating computer vision analysis, deep learning, and workflow guidance technologies, it achieves process-oriented guidance for the video acquisition process, and automates the quantification and evaluation of locomotive and train quality and personnel work quality.
[0004] According to the design scheme provided by the present invention, on the one hand, a video-based intelligent evaluation method for locomotive and train rolling stock preparation and maintenance operations is provided, comprising:
[0005] Image data of key nodes of locomotives and trains are collected, and metadata of each node is obtained based on the maintenance operation items. The metadata includes: operation time of the item, vehicle type, car number, operator, operation image of the process node and component name.
[0006] Based on the item metadata, the operational standardization of the operators and the physical condition of the locomotives and trains are inspected. The quality of the locomotives, trains and personnel is quantified and evaluated based on the inspection results.
[0007] As part of the intelligent video evaluation method for locomotive and train vehicle preparation and maintenance operations of the present invention, the method further collects image data of key nodes of locomotives and train vehicles, including:
[0008] The system uses a pre-set preparation and maintenance workflow to drive operators to collect key node image data of locomotives and trains in sequence using handheld tools. The handheld tools integrate computing power, cameras, speakers, and LED lighting. The preparation and maintenance workflow is set according to the locomotive and train models, locations, and effective times of preparation and maintenance rules.
[0009] As part of the intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations of this invention, the method further checks the level of operational standardization of personnel based on item metadata, including:
[0010] The task image in the item metadata is input into a pre-trained target detection model, and the target detection model is used to detect and identify item components in the task image. The target detection model is a deep learning model.
[0011] If the corresponding component for the item is detected and identified, the operator's work quality for that item is deemed acceptable.
[0012] If the corresponding component for a given item is not identified in the work image, the worker's work quality for that item is deemed substandard, indicating a missed inspection.
[0013] As a standardized intelligent evaluation method for locomotive preparation and maintenance operations that integrates video image acquisition, the present invention further includes checking the physical status of locomotive and train equipment based on item point metadata, comprising:
[0014] Identify the boundaries of equipment and key components inspected by maintenance personnel in each frame, and crop the key component area image from the video based on the bounding box;
[0015] Based on the item type, a target algorithm matching the item type is selected from the candidate algorithm set for quality inspection of locomotive and train rolling stock equipment. The target algorithm is used to perform anomaly checks on the component status of the item and output the check results. The candidate algorithm set for quality inspection of locomotive and train rolling stock equipment is pre-stored with execution algorithms that match each item for checking the abnormal status of the component. The execution algorithms include, but are not limited to, semantic segmentation and HSV color space analysis algorithms.
[0016] As part of the intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations of the present invention, a target algorithm matching the item type is selected to perform anomaly checks on the component status of the item and output the check results, including:
[0017] If the item point is an oil level check item point, then the semantic segmentation algorithm is used to segment the item point component region image to divide the item point component region image into oily region and oil-free region, and calculate the area of the connected region of the oily region or its proportion in the effective region of the item point component image.
[0018] The area of the connected region of the oil-containing area or its proportion in the effective area of the component image of the item point is compared with the preset corresponding threshold. If the threshold is reached, the quality of the locomotive or train vehicle at that item point is determined to be qualified and the oil level is normal. Otherwise, the quality of the locomotive or train vehicle at that item point is determined to be abnormal and the oil level is insufficient.
[0019] As part of the intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations of the present invention, a target algorithm matching the item type is selected to perform anomaly checks on the component status of the item and output the check results, including:
[0020] If the item point is a color judgment item point, then the image of the item point component region is converted from the RGB color space to the HSV color space, and the histogram of the image hue components is analyzed to determine the standard H value range corresponding to the color.
[0021] The standard H value range is used to segment the image of the component region and extract the standard color pixel region;
[0022] If the area of the segmented color standard pixel region is within the threshold range of the item, the quality of the item locomotive or train vehicle is determined to be qualified; otherwise, the quality of the item locomotive or train vehicle is determined to be abnormal. Among them, the color judgment item is the indicator item or label item.
[0023] As part of the intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations of the present invention, a target algorithm matching the item type is selected to perform anomaly checks on the component status of the item and output the check results, including:
[0024] If the item is a fastener inspection item with anti-loosening mark, the item component area image is converted from RGB color space to HSV color space, and all anti-loosening mark color areas in the item component area image are segmented according to the H value threshold corresponding to the anti-loosening mark.
[0025] Morphological processing and connectivity analysis were performed on the color regions of the anti-loosening markings, and the number of independent connected regions was counted.
[0026] If the number of connected areas is greater than the preset number of fastener anti-loosening lines, it is determined that there is a quality problem with the locomotive or train at this point, and there is a potential safety risk due to loose fasteners; if the number of connected areas is zero or less than the preset number of fastener anti-loosening lines, it is determined that there is a quality problem with the locomotive or train at this point, and there is a risk that the lack of anti-loosening lines will make it impossible to visually determine whether the fasteners are loose; if the number of connected areas is the same as the preset number of screw anti-loosening lines, it is determined that the locomotive at this point is of qualified quality and the fasteners are in normal condition.
[0027] As part of the intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations of the present invention, the method further includes, based on item metadata, checking the operational standardization of preparation and maintenance personnel and the physical condition of locomotive and train rolling stock equipment, and quantifying and evaluating locomotive quality and operational quality based on the inspection results, and also includes:
[0028] The inspection results are automatically linked to locomotive image data in time and space, and a visual report is generated.
[0029] Based on historical inspection results, the analysis results of locomotive or train rolling stock quality and operator work quality are generated according to the dimensions of work group, time period and vehicle type.
[0030] Furthermore, this invention also provides a video-based intelligent evaluation system for locomotive and train rolling stock preparation and maintenance operations, comprising: a data acquisition module and an anomaly evaluation module, wherein...
[0031] The data acquisition module is used to collect image data of key nodes of locomotives and trains, and obtain metadata of each node based on the preparation and maintenance operation items. The metadata includes: operation time of the item, vehicle type, car number, operator, operation image of the process node and component name.
[0032] The anomaly assessment module is used to check the operational standardization of preparation and maintenance personnel and the condition of locomotive equipment based on item metadata. Based on the inspection results, it automatically provides quantitative and assessment results of locomotive quality and personnel operation quality.
[0033] The beneficial effects of this invention are:
[0034] This invention integrates video image acquisition to achieve automated and quantitative assessment of locomotive and train vehicle quality and personnel work quality. It uses a workflow to guide the video acquisition process, obtaining metadata for each item in the locomotive and train vehicle preparation and maintenance work. Based on this metadata, it inspects the physical condition of the locomotive and train vehicle equipment and the level of professionalism of the preparation and maintenance personnel. The inspection results are then used to quantify and evaluate the quality of the locomotive and train vehicle and the quality of their work. This improves the efficiency, coverage, and objectivity of locomotive and train vehicle quality and work quality assessment, enhances the safety assurance capabilities of locomotives and train vehicles, and reduces the intensity of manual labor and labor costs. Attached Figure Description
[0035] Figure 1 This is a schematic diagram of the intelligent video evaluation process for locomotive and train rolling stock preparation and maintenance operations in the embodiment.
[0036] Figure 2 This is a schematic diagram of the neural network intelligent evaluation architecture for operation / locomotive quality in the embodiment;
[0037] Figure 3 This is a schematic diagram of the collaborative comprehensive judgment process for the intelligent evaluation of work / locomotive quality using a neural network, as shown in the embodiment. Detailed Implementation
[0038] To make the objectives, technical solutions, and advantages of this invention clearer and more understandable, the invention will be further described in detail below with reference to the accompanying drawings and technical solutions.
[0039] This invention addresses the issue of operation / locomotive quality assessment in intelligent operation and maintenance of rail transit. See the embodiments below. Figure 1 As shown, a video-based intelligent evaluation method for locomotive and train rolling stock preparation and maintenance operations is provided, comprising:
[0040] S101. Collect image data of key nodes of locomotives and trains, and obtain metadata of each node based on the maintenance operation items. The metadata includes: operation time of the item, vehicle type, car number, operator, operation image of the process node and component name.
[0041] Specifically, a pre-set preparation and maintenance workflow can be used to drive operators to use handheld tools to sequentially collect image data of key nodes of locomotives and trains. The handheld tools integrate computing power, cameras, speakers, and LED lighting. The preparation and maintenance workflow is set according to the locomotive and train models, locations, and effective times of preparation and maintenance rules.
[0042] like Figure 2 As shown, the intelligent camera flashlight, as a handheld tool for individual soldiers, can generate operation rules through an embedded rule configuration module. It can drive operators to perform preparation operations according to procedures in a workflow-guided manner. At preset key nodes (such as coupler inspection and front traction seat), it can automatically trigger voice prompts to guide the completion of shooting actions in a specified manner / method, ensuring the standardization and integrity of the original image data. It can also automatically segment video data of individual items based on time nodes and simultaneously generate structured metadata for each video data, including information such as operation time, process node, and component name, which is then associated and stored with the original image data.
[0043] S102. Based on item metadata, inspect the physical condition of locomotives and train rolling stock, as well as the standardization of operations by personnel. Based on the inspection results, quantify and evaluate the quality of locomotives, train rolling stock, and operational quality.
[0044] Specifically, the work image in the item metadata can be input into a pre-trained target detection model, which is a deep learning model, to detect and identify the item components in the work image. If the corresponding component of the item is detected and identified, the work quality of the operator at that item is deemed to be qualified. If the corresponding component of the item is not identified in the work image, the work quality of the operator at that item is deemed to be unqualified, indicating a missed detection.
[0045] The target detection model can employ a general deep learning-based target detection model, such as YOLOv5, to perform real-time detection on the input visual data frames, determining whether the component corresponding to the target item appears in the image. If the target item is successfully detected in any frame of the video or image sequence, the operator's work quality for that item is deemed acceptable, indicating that the inspection action has been performed. If the target item is not detected in any frame of the entire video or image sequence, the work quality for that item is deemed unacceptable, indicating that there has been a missed detection.
[0046] For items that pass the quality control test, the target region image containing the component is cropped from the corresponding frame based on the target bounding box coordinates output by the target detection model. Then, a pre-configured dedicated quality analysis algorithm matching the type of the target item or component name is scheduled for processing.
[0047] Specifically, the boundaries of the components inspected by maintenance and repair personnel can be identified from each frame of the video, and the component area image can be cropped from the video based on the bounding box. According to the type of the component, a target algorithm matching the type of the locomotive and train rolling stock equipment quality inspection is selected from the candidate algorithm set. The target algorithm is used to perform anomaly checks on the component status of the component and output the inspection results. The candidate algorithm set for locomotive and train rolling stock equipment quality inspection pre-stores execution algorithms that match each component for abnormal status checks. The execution algorithms include, but are not limited to, semantic segmentation and HSV color space analysis algorithms.
[0048] If the item is an oil level check item, the semantic segmentation algorithm is used to segment the item component region image into oil-containing and oil-free regions. The connected region area of the oil-containing region or its proportion in the effective region of the item component image is calculated. The connected region area of the oil-containing region or its proportion in the effective region of the item component image is compared with a preset threshold. If the threshold is reached, the locomotive or train vehicle at the item is determined to be of qualified quality and have a normal oil level. Otherwise, the locomotive or train vehicle at the item is determined to be of abnormal quality and have insufficient oil level.
[0049] A semantic segmentation network (such as U-Net) can be used to process the target region image, accurately segmenting the image pixels into two categories: "oil-containing areas" and "oil-free areas." The area of the connected region of the "oil-containing area" or its proportion within the effective area is calculated. This area or proportion is compared with a preset oil level area threshold for that point. Locomotive quality pass / fail determination: If the threshold is reached or exceeded, the locomotive quality at that point is determined to be pass / fail (normal oil level). Locomotive quality fail / fail determination: If the threshold is lower, the locomotive quality at that point is determined to be fail / fail (insufficient oil level).
[0050] If the item is a color judgment item, the image of the item component region is converted from the RGB color space to the HSV color space, and the histogram of the image hue components is analyzed to determine the standard H value range corresponding to the color. The image of the item component region is segmented using the standard H value range to extract the color standard pixel region. If the area of the segmented color standard pixel region is within the item threshold range, the quality of the item locomotive or train is determined to be qualified; otherwise, the quality of the item locomotive or train is determined to be abnormal. Among these, the color judgment item is an indicator item or a label item.
[0051] The target area image can be converted from the RGB color space to the HSV color space; the histogram of the image's hue (H) components is analyzed to determine the standard H value range corresponding to qualified colors; this H value range is used to perform threshold segmentation on the image to extract pixel areas whose colors meet the standard. Locomotive quality qualification judgment: If the area of the segmented color-compliant region is greater than zero (i.e., a color that meets the requirements exists), the locomotive quality at that point is judged to be qualified. Locomotive quality anomaly judgment: If no valid region is segmented, the locomotive quality at that point is judged to be abnormal (color missing or abnormal).
[0052] If the inspection point is a fastener inspection point with anti-loosening markings, the component area image of the inspection point is converted from the RGB color space to the HSV color space. All anti-loosening marking color regions in the component area image are segmented according to the H-value threshold corresponding to the anti-loosening markings. Morphological processing and connectivity analysis are performed on the anti-loosening marking color regions, and the number of independent connected regions is counted. If the number of connected regions is greater than the preset number of fastener anti-loosening lines, it is determined that there is a quality problem with the locomotive or train at this point, posing a potential safety risk due to fastener loosening. If the number of connected regions is zero or less than the preset number of fastener anti-loosening lines, it is determined that there is a quality problem with the locomotive or train at this point, posing a risk that the lack of anti-loosening lines makes it impossible to visually determine whether the fastener is loose. If the number of connected regions is the same as the preset number of screw anti-loosening lines, the locomotive at the inspection point is deemed to be of acceptable quality, and the fastener condition is normal.
[0053] The target area image can be converted to the HSV color space. A threshold H value corresponding to the red anti-loosening line is set, and all red areas in the image are segmented. Morphological processing and connectivity analysis are performed on the segmented red areas, and the number of independent red connected regions is counted. Logical judgment is made based on item knowledge: if the number of red connected regions is significantly greater than the number of anti-loosening lines the screw should have (e.g., one screw corresponds to one line, but multiple fragments are detected), it is inferred that the screw is loose, causing the anti-loosening line to break. If the number of red connected regions is zero or significantly less than the expected number, it is inferred that the anti-loosening line is missing. If the number matches the expectation and the shape is intact, the status is normal. Based on the above logic, the status of the fastener is marked as either "loose or missing" (loose or qualified locomotive quality).
[0054] like Figure 3 As shown, for the same target item, its operation quality judgment signal is correlated with the preliminary locomotive quality judgment result;
[0055] If, based on a satisfactory operational quality assessment, a "locomotive quality satisfactory" result is obtained in any frame analysis, then the final locomotive quality for that item is deemed satisfactory. If the operational quality is satisfactory, but "locomotive quality abnormal" is output in all analysis frames, then the final locomotive quality for that item is deemed abnormal. Finally, a structured inspection report containing the "operational quality" and "locomotive quality" for each item is output.
[0056] The system can also automatically link inspection results with locomotive image data in time and space, and generate a visual report; combined with historical inspection results, it can generate analysis results on locomotive or train vehicle quality and operator work quality by work group, time period and vehicle type.
[0057] The assessment results are automatically spatiotemporally linked with the original image data to generate a visual report (anomaly areas are marked with red boxes), supporting precise problem localization. Combined with historical assessment data, statistical bar charts of locomotive quality / operational quality (e.g., by shift, time period, and vehicle type) are generated to provide data support for management decisions.
[0058] Furthermore, based on the above method, this embodiment of the invention also provides a video-based intelligent evaluation system for locomotive and train rolling stock preparation and maintenance operations, comprising: a data acquisition module and an anomaly evaluation module, wherein...
[0059] The data acquisition module is used to collect image data of key nodes of locomotives and trains, and obtain metadata of each node based on the preparation and maintenance operation items. The metadata includes: operation time of the item, vehicle type, car number, operator, operation image of the process node and component name.
[0060] The anomaly assessment module is used to check the operational standardization of preparation and maintenance personnel and the condition of locomotive equipment based on item metadata. Based on the inspection results, it automatically provides quantitative and assessment results of locomotive quality and personnel operation quality.
[0061] Specifically, such as Figure 2 As shown, the data acquisition module can be a video acquisition terminal for individual soldier handheld tools, while the anomaly assessment module is a server-side module. The individual soldier handheld tool video acquisition terminal includes a workflow-guided function to drive operators to perform preparation operations according to procedures, a function to generate corresponding data for each item, and a voice and text prompt function for users during locomotive preparation operations. Users complete the work content of each item according to the guidance prompts, generating files recording each item for subsequent analysis. It also assists in rigid verification combined with work time, effectively avoiding omissions during the operation process and reducing sloppy work to a certain extent. The server-side module includes modules for setting items and rules for the individual soldier handheld tool video acquisition terminal, a work quality assessment module, a locomotive quality assessment module, a data backtracking and feedback module, and a problem chart and statistics module.
[0062] The item and rule settings module primarily provides users with the ability to configure the handheld tool acquisition terminal for individual soldiers to execute maintenance operation procedures according to changes in business operations. This includes setting methods based on locomotive and train vehicle models, locations, and the effective time of maintenance and repair rules. The configuration content is then updated to the handheld tool operation video acquisition terminal via manual or automatic synchronization.
[0063] The job quality assessment module mainly provides users with intelligent detection and identification functions for job quality. After completing the job, users can upload the work results of the single-soldier handheld tool operation video capture terminal to the server through manual upload or network upload, and the AI intelligent recognition engine will give the job quality score.
[0064] The locomotive quality assessment module mainly provides users with intelligent detection and identification functions for operational quality. After the user completes the operation, the work results of the individual soldier's handheld tool operation video acquisition terminal are uploaded to the server through manual upload or network. Based on the operation quality identification, the operation quality result is given through the AI intelligent recognition engine.
[0065] The data backtracking and feedback module mainly provides users with the functions of recognizing image results and viewing the original view operation content. After the user completes the operation, the work results of the single soldier handheld tool operation video acquisition terminal are uploaded to the server through manual upload or network. After the intelligent recognition engine of operation quality and locomotive quality gives the corresponding results, the user can view the original view data and the analyzed image data of each item.
[0066] The Problem Charts and Statistics module primarily provides users with data analysis and chart statistics functions. After the intelligent recognition engine for work quality and locomotive quality provides the corresponding results, this module investigates and analyzes the results to generate chart statistics that do not meet business requirements.
[0067] By using neural network-based intelligent assessment of work quality and locomotive quality, we can improve the efficiency, coverage, and objectivity of work quality and locomotive quality assessments, while reducing the cost and workload of manual review.
[0068] Unless otherwise specifically stated, the relative steps, numerical expressions, and values of the components and steps described in these embodiments do not limit the scope of the invention.
[0069] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.
[0070] The units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations are not considered to be beyond the scope of this invention.
[0071] Those skilled in the art will understand that all or part of the steps in the above methods can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. Optionally, all or part of the steps in the above embodiments can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiments can be implemented in hardware or as a software functional module. This invention is not limited to any particular combination of hardware and software.
[0072] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A video-based intelligent evaluation method for locomotive and train rolling stock preparation and maintenance operations, characterized in that, Include: Collect image data of key nodes of locomotives and trains, and obtain metadata of each node based on the maintenance operation items. The metadata includes: operation time of the item, vehicle type, car number, operator, operation image of the process node and component name. Based on item metadata, the physical condition of locomotives and rolling stock, as well as the degree of standardization of operations by personnel, are inspected. The quality of locomotives, rolling stock, and operations is quantified and evaluated based on the inspection results.
2. The intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations according to claim 1, characterized in that, Collect image data of key nodes of the locomotive, including: The system uses a pre-defined preparation and maintenance workflow to drive operators to sequentially collect image data of key nodes of locomotives and trains using handheld tools. The handheld tools integrate computing power, cameras, speakers, and LED lighting. The preparation and maintenance workflow is set according to the locomotive and train models, locations, and effective times of preparation and maintenance rules.
3. The intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations according to claim 1, characterized in that, The system checks the level of operational standardization of personnel at each item point based on item point metadata, including: The task image in the item metadata is input into a pre-trained target detection model, and the target detection model is used to detect and identify item components in the task image. The target detection model is a deep learning model. If the corresponding component for the item is detected and identified, the operator's work quality for that item is deemed acceptable. If the corresponding component for a given item is not identified in the work image, the worker's work quality for that item is deemed substandard, indicating a missed inspection.
4. The intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations according to claim 1 or 3, characterized in that, The status of locomotive and train rolling stock equipment is checked based on item point metadata, including: Identify the boundaries of the key components inspected by the maintenance and repair personnel in each frame, and obtain the key component area image based on the boundary cropping. Based on the item type, a target algorithm matching the item type is selected from the candidate algorithm set for quality inspection of locomotive and train rolling stock equipment. The target algorithm is used to perform anomaly checks on the component status of the item and output the check results. The candidate algorithm set for quality inspection of locomotive and train rolling stock equipment is pre-stored with execution algorithms that match each item for checking the abnormal status of the component. The execution algorithms include, but are not limited to, semantic segmentation and HSV color space analysis algorithms.
5. The intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations according to claim 4, characterized in that, Select a target algorithm that matches the item type to perform anomaly checks on the component states within the item and output the check results, including: If the item point is an oil level check item point, then the semantic segmentation algorithm is used to segment the item point component region image to divide the item point component region image into oily region and oil-free region, and calculate the area of the connected region of the oily region or its proportion in the effective region of the item point component image. The area of the connected region of the oil-containing area or its proportion in the effective area of the component image of the item point is compared with the preset corresponding threshold. If the threshold is reached, the quality of the locomotive or train vehicle at that item point is determined to be qualified and the oil level is normal. Otherwise, the quality of the locomotive or train vehicle at that item point is determined to be abnormal and the oil level is insufficient.
6. The intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations according to claim 4, characterized in that, Select a target algorithm that matches the item type to perform anomaly checks on the component states within the item and output the check results, including: If the item point is a color judgment item point, then the image of the item point component region is converted from the RGB color space to the HSV color space, and the histogram of the image hue components is analyzed to determine the standard H value range corresponding to the color. The standard H value range is used to segment the image of the component region and extract the standard color pixel region; If the area of the segmented color standard pixel region is within the threshold range of the item, the quality of the item locomotive or train vehicle is determined to be qualified; otherwise, the quality of the item locomotive or train vehicle is determined to be abnormal. Among them, the color judgment item is the indicator item or label item.
7. The intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations according to claim 4, characterized in that, Select a target algorithm that matches the item type to perform anomaly checks on the component states within the item and output the check results, including: If the item is a fastener inspection item with anti-loosening mark, the item component area image is converted from RGB color space to HSV color space, and all anti-loosening mark color areas in the item component area image are segmented according to the H value threshold corresponding to the anti-loosening mark. Morphological processing and connectivity analysis were performed on the color regions of the anti-loosening markings, and the number of independent connected regions was counted. If the number of connected areas is greater than the number of preset fastener anti-loosening lines, it is determined that there is a quality problem with the locomotive or train at this point, and there is a potential safety risk due to loose fasteners. If the number of connected areas is zero or less than the preset number of fastener anti-loosening lines, it is determined that there is a quality problem with the locomotive or train at this point, and there is a risk that the lack of anti-loosening lines will make it impossible to intuitively judge whether the fasteners are loose; if the number of connected areas is the same as the preset number of screw anti-loosening lines, it is determined that the locomotive at this point is of qualified quality and the fasteners are in normal condition.
8. The intelligent video evaluation method for locomotive and train rolling stock preparation and maintenance operations according to claim 1, characterized in that, Based on itemized metadata, the inspection assesses the operational standards of maintenance and repair personnel, as well as the physical condition of locomotives and rolling stock. The inspection results are used to quantify and evaluate locomotive quality and operational quality. This also includes: The inspection results are automatically linked to locomotive image data in time and space and a visual report is generated. Based on historical inspection results, the analysis results of locomotive or train rolling stock quality and operator work quality are generated according to the dimensions of work group, time period and vehicle type.
9. A video-based intelligent evaluation system for locomotive and train rolling stock preparation and maintenance operations, characterized in that, Include: The system includes a data acquisition module and an anomaly assessment module. The data acquisition module is used to collect image data of key nodes of locomotives and trains, and obtain metadata of each node based on the preparation and maintenance operation items. The metadata includes: operation time of the item, vehicle type, car number, operator, operation image of the process node and component name. The anomaly assessment module is used to check the operational standardization of preparation and maintenance personnel and the condition of locomotive equipment based on item metadata. Based on the inspection results, it automatically provides quantitative and assessment results of locomotive quality and personnel operation quality.
10. An electronic device, characterized in that, include: At least one CPU processor, one GPU or NPU processor, and a memory coupled to the processor; wherein the memory stores a front-end application and an AI model, the front-end application and the AI model being executable by the processor to implement the method as described in any one of claims 1 to 8.