Visual detection method and system for standardability of wall-tapping and top-asking process

Abstract

The invention relates to a normative visual detection method for a wall-knocking and top-asking process, which belongs to the field of safety monitoring and comprises the following steps: (1) airborne support or temporary support mark detection: identifying an airborne support or temporary support mark in a roadway through a target detection algorithm, and judging whether support is carried out or not; (2) detection of wall-knocking and top-asking personnel: after the support mark is detected, analyzing whether the personnel in the roadway meet the requirements of wall-knocking and top-asking; (3) wall-knocking and top-asking action detection: monitoring actions of wall-knocking and top-asking personnel, and judging whether standard wall-knocking and top-asking operation is performed or not; and (4) position detection: ensuring that the upper-knocking and top-asking actions occur at the left position, the middle position and the right position of the monitoring picture so as to judge the normalization of the process. The coal mine safety monitoring system has the advantages that safety monitoring efficiency is improved, whether wall knocking and top asking are carried out or not is automatically identified through 24-hour all-weather dead-corner-free real-time monitoring, the safety monitoring efficiency of a coal mine is remarkably improved, and the response speed, the processing efficiency and the like are improved.

Description

Technical Field

[0001] The invention relates to a standardized visual detection method for knocking and top-asking procedures, and belongs to the technical field of industrial safety monitoring. Background Art

[0002] In the process of coal mining, the system of knocking the roof and asking about the top must be strictly implemented before the roadway support is carried out. Since construction and production will lead to stress redistribution of the newly exposed roof, coal wall and two sides, this process may cause the surrounding coal rock to gradually separate from the main body. Especially in the presence of bedding, cracks or complex geological structures, the risk of roof collapse and spalling accidents increases significantly, which may cause serious casualties.

[0003] Therefore, it is very important to conduct a comprehensive safety inspection of the working face before supporting. Each team member should promptly check the roof and the side of their work area and deal with any potential dangers to ensure the safety of the working face and prevent accidents. However, the existing technology has the following disadvantages in the standardized inspection of the side and top process:

[0004] High dependence on manual work: Current detection methods mainly rely on manual inspection, which is easily affected by subjective factors, resulting in inaccurate judgments and increased safety risks.

[0005] Inefficiency: Manual inspection takes a lot of time and manpower, especially in large-scale coal mining operations, which makes it difficult to achieve real-time monitoring and rapid response.

[0006] Information is not recorded in a timely manner: Manual recording often has delays and cannot provide timely feedback on potential safety issues, increasing the risk of accidents.

[0007] Lack of systematic monitoring: The existing methods lack a comprehensive monitoring system, making it difficult to achieve real-time monitoring and data analysis of the entire working surface, and unable to effectively identify and warn of safety hazards. Summary of the invention

[0008] In order to overcome the defects of the prior art, the present invention provides a standardized visual inspection method for the knocking and top-asking process. The technical solution of the present invention is:

[0009] A standardized visual inspection method for a knocking and top-asking process comprises the following steps:

[0010] (1) Detection of airborne support or temporary support signs: The target detection algorithm is used to identify airborne support or temporary support signs in the tunnel to determine whether support has been carried out;

[0011] (2) Personnel detection: After the support mark is detected, analyze whether the personnel in the tunnel meet the requirements of knocking and checking the top, including the number of people and whether they have tools;

[0012] (3) Detection of knocking action: monitoring the actions of the knocking personnel to determine whether the knocking operation is carried out in a standardized manner;

[0013] (4) Position detection: Ensure that the knocking action occurs at the left, middle, and right positions of the monitoring screen to determine the standardization of the process.

[0014] The step (1) is specifically as follows:

[0015] 1.1 Data collection: Use the camera to collect video stream data of the working surface in real time;

[0016] 1.2 Video preprocessing: preprocess the collected video stream, including frame rate control, noise filtering and image enhancement, to improve the accuracy of subsequent target detection;

[0017] 1.3 Target detection algorithm selection: Use YOLOX algorithm for recognition and detection;

[0018] 1.4 Model training: Use the labeled training data set to train the YOLOX model; the training data set contains images of airborne support and temporary support signs in various working face conditions; 1.5 Model deployment: Deploy the trained YOLOX model to the edge analysis device;

[0019] 1.6 Real-time detection: Use the YOLOX model to analyze the real-time video stream captured by the camera to identify whether there are signs of airborne support or temporary support.

[0020] 1.7 Sign recognition: For each frame of the image, the YOLOX model outputs bounding boxes that identify the target objects in the image, as well as the corresponding confidence scores; by setting a confidence threshold, when the confidence is higher than the threshold, the detected sign is considered valid; 1.8 Result judgment: If the signs of airborne support or temporary support are continuously detected in consecutive video frames, it is considered that the working face has been supported; if no signs are detected within the preset time, it is considered that the working face has not been supported;

[0021] 1.9 Alarm mechanism: When it is determined that the working face is not supported, the alarm mechanism will be triggered, and the relevant personnel will be reminded to take measures through sound and light alarm or sending notification to the monitoring center.

[0022] 1.10 Evidence video generation: Save the video clips containing undetected support signs as evidence videos for subsequent analysis and processing.

[0023] The step (2) is specifically as follows:

[0024] 2.1 Video frame analysis: After detecting signs of machine-borne support or temporary support, use the target detection algorithm to analyze the backtracking video frames to identify and locate personnel in the tunnel; 2.2 Personnel detection: Identify all personnel in the video frame, identify whether the personnel are holding hand picks, and determine whether they are personnel performing knocking and top-checking tasks;

[0025] 2.3 Number of people and role judgment: Based on the number of people detected and their roles, the system will make the following judgments:

[0026] If there is no one in the retracing video, or there is only one person and no one is involved in the knocking and top-checking process, it is considered that the knocking and top-checking process has not been carried out, and an alarm is triggered;

[0027] If only the knocker is detected without a guardian, or if there are more than two knockers, the knocking process is considered irregular and an alarm is triggered. If two persons are detected, including the knocker and the guardian, the next step is to detect the knocker's movements.

[0028] 2.4 Action standardization assessment: Analyze whether the actions of the personnel knocking on the roof comply with the prescribed operating procedures, such as whether the hand pick is used to knock on the roof or coal wall in the correct position.

[0029] If the action does not meet the specifications, the event will be recorded and may trigger an alarm to alert the on-site safety manager to intervene;

[0030] 2.5 Alarm and Recording: If irregular operation or potential safety risk is detected, an alarm will be generated and the relevant video clips and alarm information will be uploaded to the central monitoring system.

[0031] The step (3) is specifically as follows:

[0032] 3.1 Motion detection preparation: After confirming that there are a person who knocks on the top and a guardian present, the person's movements will be analyzed;

[0033] 3.2 Video frame extraction: Extract consecutive frames containing gang members from the surveillance video. These frames are used for subsequent action recognition analysis.

[0034] 3.3 Action recognition: Use the X-CLIP algorithm to analyze the extracted video frames to identify the specific actions of the gang members;

[0035] 3.4 Standardization judgment: If the action of knocking the top plate or side panels cannot be identified, the knocking process is considered to be non-standard and an alarm is triggered; if a knocking action is detected, position detection is performed.

[0036] The step (4) is specifically as follows:

[0037] 4.1 Division of monitoring screen: The monitoring screen is evenly divided into three areas: left, middle and right, so as to accurately identify the position of the knocking action;

[0038] 4.2 Video frame extraction: Extract video frames containing knocking and asking actions from the surveillance video. These frames will be used for subsequent position detection analysis;

[0039] 4.3 Application of target detection algorithm: Use YOLOX or other target detection algorithms to analyze the extracted video frames to identify the action positions of the gang members;

[0040] 4.4 Position standardization judgment: Detect whether the knocking action occurs in the three predetermined areas of the left, middle and right of the monitoring screen; if the knocking action does not occur in these three areas, it is considered that the knocking process is not standardized and an alarm is triggered;

[0041] 4.5 Action location record: For each detected knocking action, record its specific location in the monitoring screen to facilitate subsequent analysis and review;

[0042] 4.6 Alarm and feedback: If the knocking and top-asking process is detected to be irregular, an alarm will be triggered and the relevant video clips and alarm information will be uploaded to the central monitoring system; at the same time, the feedback results will be given to the on-site operators;

[0043] 4.7 Data storage and analysis: All detected motion position data will be stored in the database for subsequent data analysis and trend prediction.

[0044] A system for realizing a standardized visual inspection method for knocking and top-asking processes, comprising:

[0045] An airborne support or temporary support mark detection module is used to identify airborne support or temporary support marks in the tunnel through a target detection algorithm to determine whether support has been performed;

[0046] The personnel detection module for knocking on the side and asking about the top is used to analyze whether the personnel in the tunnel meet the requirements for knocking on the side and asking about the top after detecting the support mark;

[0047] The knocking action detection module is used to monitor the actions of the knocking personnel and determine whether the knocking operation is carried out in a standardized manner;

[0048] The position detection module is used to ensure that the knocking action occurs at the left, middle and right positions of the monitoring screen to determine the standardization of the process.

[0049] The advantages of the present invention are:

[0050] 1. Improve safety monitoring efficiency: Through 24-hour all-weather real-time monitoring without blind spots and automatic identification of whether to knock on the side or top, it not only significantly improves the safety monitoring efficiency of coal mines, but also improves the response speed and processing efficiency.

[0051] 2. Reduce labor intensity: greatly reduce the traditional "man-to-man" supervision model and reduce the burden on workers.

[0052] 3. Enhanced safety: Through real-time monitoring and intelligent analysis, it is possible to promptly discover processes that have not been carried out, thereby enhancing the overall safety of coal mining.

[0053] The present invention monitors the movements and number of personnel in the tunnel in real time, as well as the signs of temporary support or airborne support, and retrospectively checks the video of the previous thirty minutes of temporary support or airborne support to determine whether a knocking operation has occurred, and forms an evidence video for alarm. The advantage of this standardization detection method, equipment and system for knocking and top-questioning procedures based on visual analysis is that it can monitor the situation in the tunnel in real time, and determine whether a knocking and top-questioning operation has occurred by retrospective means. Compared with the traditional method of looking back at the camera, the present invention can more accurately determine the occurrence of a knocking and top-questioning operation, and can handle it in a timely manner, thereby improving safety and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic flow chart of the method of the present invention.

[0055] Figure 2 yes Figure 1 Schematic diagram of the process of personnel detection in .

[0056] Figure 3 It is a system structure block diagram of the present invention. DETAILED DESCRIPTION

[0057] The present invention will be further described below in conjunction with specific embodiments, and the advantages and features of the present invention will become clearer as the description proceeds. However, these embodiments are exemplary only and do not constitute any limitation to the scope of the present invention. It should be understood by those skilled in the art that the details and forms of the technical solution of the present invention may be modified or replaced without departing from the spirit and scope of the present invention, but these modifications and replacements all fall within the scope of protection of the present invention.

[0058] See also Figures 1 to 3 The present invention relates to a standardized visual inspection method for knocking and top-asking processes, comprising the following steps:

[0059] (1) Detection of airborne support or temporary support signs: The target detection algorithm is used to identify airborne support or temporary support signs in the tunnel to determine whether support has been carried out;

[0060] (2) Personnel detection: After the support mark is detected, analyze whether the personnel in the tunnel meet the requirements of knocking and checking the top, including the number of people and whether they have tools;

[0061] (3) Detection of knocking action: monitoring the actions of the knocking personnel to determine whether the knocking operation is carried out in a standardized manner;

[0062] (4) Position detection: Ensure that the knocking action occurs at the left, middle, and right positions of the monitoring screen to determine the standardization of the process.

[0063] The standardized visual inspection method for the knocking and top-asking process of the present invention has the following advantages:

[0064] High degree of automation: Through automated visual inspection methods, the reliance on manual inspection is reduced, and the efficiency and accuracy of inspection are improved.

[0065] Real-time monitoring: It can monitor the knocking and top-questioning processes in the tunnel in real time and discover potential safety hazards in time.

[0066] Comprehensiveness: Not only the support signs are detected, but also the personnel, actions and positions are fully detected, ensuring the comprehensiveness of the detection.

[0067] Accuracy: Using advanced target detection algorithms, such as YOLOX, the recognition accuracy of airborne support or temporary support signs is improved.

[0068] Improved safety: By timely identifying irregular roof knocking and roof questioning operations, safety accidents such as roof collapse and roof spalling can be effectively prevented.

[0069] Data recording: The system can automatically record test results and related video evidence to facilitate post-analysis and responsibility tracing.

[0070] The step (1) is specifically as follows:

[0071] 1.1 Data collection: Use the camera to collect video stream data of the working surface in real time;

[0072] 1.2 Video preprocessing: preprocess the collected video stream, including frame rate control, noise filtering and image enhancement, to improve the accuracy of subsequent target detection;

[0073] 1.3 Target detection algorithm selection: Use YOLOX algorithm for recognition and detection;

[0074] 1.4 Model training: Use the labeled training data set to train the YOLOX model; the training data set contains images of airborne support and temporary support signs in various working face conditions; 1.5 Model deployment: Deploy the trained YOLOX model to edge analysis devices; such as inference servers or edge analysis boxes with AI chips;

[0075] 1.6 Real-time detection: Use the YOLOX model to analyze the real-time video stream captured by the camera to identify whether there are signs of airborne support or temporary support.

[0076] 1.7 Sign recognition: For each frame of the image, the YOLOX model outputs bounding boxes that identify the target objects in the image, as well as the corresponding confidence scores; by setting a confidence threshold, when the confidence is higher than the threshold, the detected sign is considered valid; 1.8 Result judgment: If the signs of airborne support or temporary support are continuously detected in consecutive video frames, it is considered that the working face has been supported; if no signs are detected within the preset time, it is considered that the working face has not been supported;

[0077] 1.9 Alarm mechanism: When it is determined that the working face is not supported, the alarm mechanism will be triggered, and the relevant personnel will be reminded to take measures through sound and light alarm or sending notification to the monitoring center.

[0078] 1.10 Evidence video generation: Save the video clips containing undetected support signs as evidence videos for subsequent analysis and processing.

[0079] This step (1) achieves:

[0080] Efficient data collection: Real-time video stream data collection ensures the continuity and real-time nature of detection, providing sufficient information for subsequent analysis.

[0081] Video preprocessing optimization: The preprocessing step improves image quality, reduces the interference of environmental noise on detection results, and enhances the accuracy of target detection.

[0082] Advanced target detection algorithm: Uses the YOLOX algorithm, which is an efficient target detection algorithm that can quickly and accurately identify airborne support or temporary support signs.

[0083] Reduced manual reliance: Reduced reliance on manual inspections, lowered labor costs, and avoided detection oversights caused by human factors.

[0084] Improve safety management level: Through automated visual inspection, the safety management level of underground projects such as coal mines is improved, which helps to achieve safe production.

[0085] The step (2) is specifically as follows:

[0086] 2.1 Video frame analysis: After detecting signs of machine-borne support or temporary support, use the target detection algorithm to analyze the backtracking video frames to identify and locate personnel in the tunnel; 2.2 Personnel detection: Identify all personnel in the video frame, identify whether the personnel are holding hand picks, and determine whether they are personnel performing knocking and top-checking tasks;

[0087] 2.3 Number of people and role judgment: Based on the number of people detected and their roles (criminals and guardians), the following judgments will be made:

[0088] If there is no one in the retracing video, or there is only one person and no one is involved in the knocking and top-checking process, it is considered that the knocking and top-checking process has not been carried out, and an alarm is triggered;

[0089] If only the person who knocks on the roof is detected and there is no guardian present, or there are more than two people knocking on the roof, it is considered that the knocking on the roof process is not standardized and the alarm is also triggered;

[0090] If two persons are detected, including the person who knocks on the door and the guardian, the next step will be to detect the action of the person who knocks on the door;

[0091] 2.4 Action standardization assessment: Analyze whether the actions of the personnel knocking on the roof comply with the prescribed operating procedures, such as whether the hand pick is used to knock on the roof or coal wall in the correct position.

[0092] If the action does not meet the specifications, the event will be recorded and may trigger an alarm to alert the on-site safety manager to intervene;

[0093] 2.5 Alarm and Recording: If irregular operation or potential safety risk is detected, an alarm will be generated and the relevant video clips and alarm information will be uploaded to the central monitoring system.

[0094] Advantages of this step (2):

[0095] Comprehensive video frame analysis: Through the analysis of retrospective video frames, the system can accurately identify people in the lanes, ensuring that every work link is monitored without blind spots.

[0096] Efficient Person Detection: Using the target detection algorithm, all persons in the video frame and the tools they hold (such as hand picks) can be quickly identified, improving the accuracy and efficiency of detection.

[0097] Intelligent role judgment: The system automatically judges the roles of personnel (criminal and guardian), reducing the need for manual intervention and the possibility of human error.

[0098] Timely alarm mechanism: If it is detected that the knocking and top-asking process is not carried out or irregular operations are performed, the system can quickly trigger an alarm to remind on-site staff in time, thereby enhancing safety.

[0099] The step (3) is specifically as follows:

[0100] 3.1 Motion detection preparation: After confirming that there are a person who knocks on the top and a guardian present, the person's movements will be analyzed;

[0101] 3.2 Video frame extraction: Extract consecutive frames containing gang members from the surveillance video. These frames are used for subsequent action recognition analysis.

[0102] 3.3 Action recognition: Use the X-CLIP algorithm to analyze the extracted video frames to identify the specific actions of the gang members;

[0103] 3.4 Standardization judgment: If the action of knocking the top plate or side panels cannot be identified, the knocking process is considered to be non-standard and an alarm is triggered; if a knocking action is detected, position detection is performed.

[0104] Advantages of this step (3):

[0105] Accurate action recognition: By using the X-CLIP algorithm, the system can accurately identify the specific actions of people knocking on the roof, including the behavior of knocking on the roof or side panels.

[0106] Real-time motion analysis: The extraction and analysis of continuous frames ensures the real-time nature of motion detection, enabling the system to respond and handle irregular operations in a timely manner.

[0107] Automatic alarm mechanism: When the system identifies irregular movements or the lack of necessary knocking movements, it can automatically trigger an alarm, improving response speed and safety.

[0108] Reduce false alarms: Accurate motion recognition reduces false alarms caused by misjudgment, ensuring the effectiveness and reliability of the alarm system.

[0109] Improve work efficiency: Automated motion detection reduces the workload of manual monitoring, allowing safety supervisors to focus on tasks that require more manual judgment.

[0110] Adapt to complex environments: The advanced nature of the X-CLIP algorithm enables the system to adapt to different lighting conditions and working environments, ensuring accurate action recognition under various conditions.

[0111] The step (4) is specifically as follows:

[0112] 4.1 Division of monitoring screen: The monitoring screen is evenly divided into three areas: left, middle and right, so as to accurately identify the position of the knocking action;

[0113] 4.2 Video frame extraction: Extract video frames containing knocking and asking actions from the surveillance video. These frames will be used for subsequent position detection analysis;

[0114] 4.3 Application of target detection algorithm: Use YOLOX or other target detection algorithms to analyze the extracted video frames to identify the action positions of the gang members;

[0115] 4.4 Position standardization judgment: Detect whether the knocking action occurs in the three predetermined areas of the left, middle and right of the monitoring screen; if the knocking action does not occur in these three areas, it is considered that the knocking process is not standardized and an alarm is triggered; if the knocking action occurs in the left, middle and right positions of the monitoring screen, it is considered that the knocking process is standardized;

[0116] 4.5 Action location record: For each detected knocking action, record its specific location in the monitoring screen to facilitate subsequent analysis and review;

[0117] 4.6 Alarm and feedback: If the knocking and top-asking process is detected to be irregular, an alarm will be triggered and the relevant video clips and alarm information will be uploaded to the central monitoring system; at the same time, the feedback results will be given to the on-site operators;

[0118] 4.7 Data storage and analysis: All detected motion position data will be stored in the database for subsequent data analysis and trend prediction.

[0119] This step (4) achieves:

[0120] Accurate position monitoring: By dividing the monitoring screen into three areas: left, center, and right, the system can accurately identify whether the knocking action occurs in the correct position.

[0121] Efficient video processing: Extraction and analysis of video frames ensures real-time monitoring and rapid response to tapping and asking actions.

[0122] Advanced target detection: The use of advanced algorithms such as YOLOX improves the accuracy of action location recognition and ensures the reliability of detection results.

[0123] Automatic judgment of standardization: The system can automatically judge whether the knocking and top-asking actions are in compliance with the standards, reducing the need for manual monitoring and improving efficiency.

[0124] The present invention also relates to a system for realizing a standardized visual inspection method for knocking and top-asking processes, comprising:

[0125] An airborne support or temporary support mark detection module 1 is used to identify airborne support or temporary support marks in the tunnel through a target detection algorithm to determine whether support has been performed;

[0126] The personnel detection module 2 for knocking on the sidewalls and asking about the top is used to analyze whether the personnel in the tunnel meet the requirements for knocking on the sidewalls and asking about the top after detecting the support mark;

[0127] The knocking action detection module 3 is used to monitor the actions of the knocking personnel and determine whether the knocking operation is performed in a standardized manner;

[0128] The position detection module 4 is used to ensure that the knocking action occurs at the left, middle and right positions of the monitoring screen to determine the standardization of the process.

[0129] The working principle of the standardized visual inspection method for the knocking and top-asking process of the present invention is as follows:

[0130] Airborne support or temporary support mark detection:

[0131] Use the camera to collect video stream data of the working surface in real time.

[0132] The video stream is preprocessed, including frame rate control, noise filtering, and image enhancement, to improve the accuracy of subsequent target detection.

[0133] Select a suitable target detection algorithm (such as YOLOX) for recognition and detection.

[0134] The model is trained using a labeled training dataset containing images of airborne support and temporary support signs in various working face conditions.

[0135] Deploy the trained model to edge analysis devices, such as inference servers with AI chips or edge analysis boxes.

[0136] The video stream captured by the camera is detected in real time to identify whether there are signs of airborne support or temporary support.

[0137] Testing of personnel who ask for help:

[0138] After the support signs are detected, the backtracked video frames are analyzed using an object detection algorithm to identify and locate the personnel in the roadway.

[0139] Identify all persons in the video frame, identify whether the person is holding a hand pickaxe tool, and determine whether he is the person performing the knocking task.

[0140] Based on the number of people detected and their roles (criminals and guardians), a judgment is made and the corresponding alarm is triggered.

[0141] Knock-and-ask action detection:

[0142] After confirming that there are a gang member and a guardian present, analyze the actions of the gang member.

[0143] Continuous frames containing gang members are extracted from surveillance videos for subsequent action recognition analysis.

[0144] The extracted video frames are analyzed using the X-CLIP algorithm to identify the specific actions of the criminals.

[0145] If the action of knocking the top plate or side panels is not identified, it is considered that the knocking process is not standardized and an alarm is triggered.

[0146] Position detection:

[0147] The monitoring screen is evenly divided into three areas: left, middle and right, so as to accurately identify the location of the knocking action.

[0148] Video frames containing knocking and asking actions are extracted from the surveillance video for subsequent position detection analysis.

[0149] The extracted video frames are analyzed using an object detection algorithm to identify the locations of the actions of the attackers.

[0150] Detect whether the knocking and top-asking actions occur in the three predetermined areas of the left, middle and right of the monitoring screen, and trigger an alarm or confirm the standardization of the process based on the results.

[0151] The modules included in the system of the present invention can automatically detect the standardization of the knocking and top-questioning process through the coordinated work of these modules, and timely discover and prompt irregular operations, thereby improving the safety and efficiency of coal mine operations.

[0152] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A standardized visual inspection method for knocking and top-asking process, characterized in that: The following steps are involved: (1) Detection of airborne support or temporary support signs: The target detection algorithm is used to identify airborne support or temporary support signs in the tunnel to determine whether support has been carried out; (2) Personnel detection: After the support mark is detected, analyze whether the personnel in the tunnel meet the requirements of knocking and checking the top, including the number of people and whether they have tools; (3) Detection of knocking action: monitoring the actions of the knocking personnel to determine whether the knocking operation is carried out in a standardized manner; (4) Position detection: Ensure that the knocking action occurs at the left, middle, and right positions of the monitoring screen to determine the standardization of the process.

2. The method for standardizing visual inspection of knocking and top-asking process according to claim 1 is characterized in that: The step (1) is specifically as follows: 1.1 Data collection: Use the camera to collect video stream data of the working surface in real time; 1.2 Video preprocessing: preprocess the collected video stream, including frame rate control, noise filtering and image enhancement, to improve the accuracy of subsequent target detection; 1.3 Target detection algorithm selection: Use YOLOX algorithm for recognition and detection; 1.4 Model training: The YOLOX model is trained using a labeled training dataset; the training dataset contains images of airborne support and temporary support signs under various working face conditions; 1.5 Model deployment: Deploy the trained YOLOX model to the edge analysis device; 1.6 Real-time detection: Use the YOLOX model to analyze the real-time video stream captured by the camera to identify whether there are signs of airborne support or temporary support. 1.7 Landmark Recognition: For each frame of the image, the YOLOX model outputs bounding boxes that identify the target objects in the image and the corresponding confidence scores; by setting a confidence threshold, when the confidence is higher than the threshold, the detected landmark is considered valid; 1.8 Result judgment: If the signs of airborne support or temporary support are continuously detected in consecutive video frames, it is considered that the working face has been supported; if no signs are detected within the preset time, it is considered that the working face is not supported. 1.9 Alarm mechanism: When it is determined that the working face is not supported, the alarm mechanism will be triggered, and the relevant personnel will be reminded to take measures through sound and light alarm or sending notification to the monitoring center. 1.10 Evidence video generation: Save the video clips containing undetected support signs as evidence videos for subsequent analysis and processing.

3. The method for standardizing visual inspection of the knocking and top-questioning process according to claim 1 or 2 is characterized in that: The step (2) is specifically as follows: 2.1 Video frame analysis: After detecting signs of airborne support or temporary support, the target detection algorithm is used to analyze the retrospective video frames to identify and locate the personnel in the tunnel; 2.2 Personnel detection: Identify all persons in the video frame, identify whether the person is holding a pickaxe, and determine whether the person is performing the task of knocking down the gang; 2.3 Number of people and role judgment: Based on the number of people detected and their roles, the following judgments are made: If there is no one in the retracing video, or there is only one person and no one is involved in the knocking and top-checking process, it is considered that the knocking and top-checking process has not been carried out, and an alarm is triggered; If only the person who knocks on the roof is detected and there is no guardian present, or there are more than two people knocking on the roof, it is considered that the knocking on the roof process is not standardized and the alarm is also triggered; If two persons are detected, including the person who is asking for trouble and the guardian, the process will proceed to the next step, which is to detect the movements of the person who is asking for trouble.

4. The method for standardizing visual inspection of the knocking and top-questioning process according to claim 1 or 2 is characterized in that: The step (3) is specifically as follows: 3.1 Motion detection preparation: After confirming that there are a gang-banging person and a guardian present, the movements of the gang-banging person will be analyzed; 3.2 Video frame extraction: Extract consecutive frames containing gang members from the surveillance video. These frames are used for subsequent action recognition analysis. 3.3 Action recognition: Use the X-CLIP algorithm to analyze the extracted video frames to identify the specific actions of the gang members; 3.4 Standardization judgment: If the action of knocking the top plate or side panels cannot be identified, the knocking process is considered to be non-standard and an alarm is triggered; if a knocking action is detected, position detection is performed.

5. The method for standardizing visual inspection of the knocking and top-questioning process according to claim 1 or 2 is characterized in that: The step (4) is specifically as follows: 4.1 Division of monitoring screen: The monitoring screen is evenly divided into three areas: left, middle and right, so as to accurately identify the position of the knocking action; 4.2 Video frame extraction: Extract video frames containing knocking and asking actions from the surveillance video. These frames will be used for subsequent position detection analysis; 4.3 Application of target detection algorithm: Use YOLOX algorithm to analyze the extracted video frames to identify the action positions of the gang members; 4.4 Position standardization judgment: Detect whether the knocking action occurs in the three predetermined areas of the left, middle and right of the monitoring screen; if the knocking action does not occur in these three areas, it is considered that the knocking process is not standardized and an alarm is triggered; 4.5 Action location record: For each detected knocking action, record its specific location in the monitoring screen to facilitate subsequent analysis and review; 4.6 Alarm and feedback: If the knocking and top-asking process is detected to be irregular, an alarm will be triggered and the relevant video clips and alarm information will be uploaded to the central monitoring system; at the same time, the feedback results will be given to the on-site operators; 4.7 Data storage and analysis: All detected motion position data will be stored in the database for subsequent data analysis and trend prediction.

6. A system for implementing the standardized visual inspection method for knocking and top-checking process according to any one of claims 1 to 5, characterized in that: include: An airborne support or temporary support mark detection module is used to identify airborne support or temporary support marks in the tunnel through a target detection algorithm to determine whether support has been performed; The personnel detection module for knocking on the side and asking about the top is used to analyze whether the personnel in the tunnel meet the requirements of knocking on the side and asking about the top after detecting the support mark; The knocking action detection module is used to monitor the actions of the knocking personnel and determine whether the knocking operation is carried out in a standardized manner; The position detection module is used to ensure that the knocking action occurs at the left, middle and right positions of the monitoring screen to determine the standardization of the process.

Images