A tile inspection operation compliance identification method based on a target tracking algorithm

Abstract

The application discloses a kind of based on target tracking algorithm's gas detection operation compliance identification method, belong to gas detection technical field.The method includes: high-definition video camera is arranged in the methane detection area of mine, real-time acquisition video data in methane detection area and image is preprocessed;By YOLO target detection method, the gas detector in the methane detection area in video is identified;If detecting that gas detector hand holds gas detector and arrives specified methane detection area on time, then it is judged as compliance operation, otherwise it is judged as non-compliance operation;After verifying that gas detector hand holds gas detector arrives specified methane detection area on time, the gas detection operation of scene is tracked in real time by CSRT target tracking algorithm;Then with standardized operation is compared, if gas detection operation meets standardized operation, then it is judged as compliance operation, otherwise it is judged as non-compliance operation.The application can improve the reliability and safety of gas detection operation, reduce the safety risk brought by illegal operation or non-operation.

Description

Technical Field

[0001] This invention relates to the field of coal mine safety monitoring and intelligent video analysis technology, specifically a method for compliance identification of gas inspection operations based on target tracking algorithms. It is particularly suitable for complex lighting environments underground, enabling automated supervision of gas inspector behavior trajectory tracking, standardized action sequence verification, and real-time alarm for violations. It belongs to the cross-application field of intelligent mine safety monitoring and industrial automation technology. Background Technology

[0002] Methane gas is a serious threat to coal mine safety, causing accidents such as combustion, explosion, and poisoning, resulting in casualties and economic losses. Methane detection is a core aspect of coal mine safety. Traditional methods rely on inspectors manually carrying portable detectors to measure gas concentrations along patrol routes. Currently, manual inspections suffer from strong subjectivity and incomplete oversight, making it difficult to prevent missed detections (missing key areas) and falsified data. Underground production systems are vast and the environment is complex and variable, creating blind spots in gas detection supervision. According to statistics from the National Mine Safety Administration, many gas detection operations in methane accidents involve non-standard practices. Therefore, effectively supervising gas detection operations and ensuring compliance is a problem that needs to be solved in this technical field. Summary of the Invention

[0003] To address the issues of missed detections and false detections in existing tile inspection operations, this invention provides a compliance identification method for tile inspection operations based on a target tracking algorithm. This method uses computer vision technology to intelligently monitor the entire tile inspection process, effectively improving the standardization of inspection operations and the reliability of data. The specific technical solution is as follows:

[0004] To address the issues of missed and false inspections in gas inspection operations, this invention discloses a compliance identification method for gas inspection operations based on a target tracking algorithm. This method uses an intelligent video monitoring system to track the behavior and actions of gas inspectors in real time, automatically identifying these actions and comparing their behavior patterns with standard operating procedures to confirm whether each operation complies with predetermined safety standards. Alarms are issued for detected violations to remind inspectors to correct them. This method can significantly reduce the safety risks caused by human error or intentional violations.

[0005] The technical solution adopted in this invention is: a compliance identification method for tile inspection operations based on a target tracking algorithm, comprising the following steps:

[0006] Step S1: Deploy a high-definition video camera in the methane detection area of ​​the mine to collect video data in the methane detection area in real time and preprocess the images.

[0007] Step S2: Using the YOLO target detection algorithm, identify the gas inspector and gas detector in the methane detection area in the video; if the gas inspector is detected holding a gas detector and arrives at the designated methane detection area on time, it is determined to be a compliant operation; otherwise, it is determined to be a non-compliant operation.

[0008] Step S3: After verifying that the gas inspector arrives at the designated methane detection area on time with a gas detector, the CSRT target tracking algorithm is used to track the gas inspection operation in real time. Then, the operation is compared with the established standardized procedures. If the gas inspection operation conforms to the standardized procedures, it is considered a compliant operation; otherwise, it is considered a non-compliant operation. The standardized procedures include:

[0009] (1) Single gas detection procedure: Hold the gas detector high, squeeze the balloon and lower your head to read the value;

[0010] (2) Repeat the single detection action process according to the set number of times.

[0011] As a further improvement of the present invention, in step S1, the high-definition video camera has a wide dynamic range function, and the light intensity range inside the mine is set to I. min To I max The selected camera's wide dynamic range (DR) meets the requirements.

[0012] As a further improvement of the present invention, in step S2, the method for determining whether the gas inspector has arrived on time is to analyze the arrival time T of the gas inspector through video image analysis. arrive With the specified arrival time T required When T arrive >T required At that time, it was determined that the inspection operation was not in compliance with regulations.

[0013] As a further improvement of the present invention, in step S3, the method for comparing on-site tile inspection operations with standardized operations is as follows:

[0014] (1) Action determination of the high-lift gas detector: Let the methane detection area be R = (x min ,y min ,x max ,y max When the gas detector is raised to the range (x, y) and x satisfies min ≤x≤x max and y min ≤y≤y max At that time, the execution of the action is confirmed through target tracking;

[0015] (2) Determining the action of squeezing the balloon: A target tracking algorithm is used to monitor the inspector's hand movements. The initial coordinates of the key hand points are set as follows: The position coordinates in subsequent frames are By calculating the Euclidean distance When d t The action of squeezing the balloon by hand is considered valid when the number of times the squeezing distance is within the preset range and the preset value N is met consecutively.

[0016] (3) Determining the action of looking down to read: Let the initial value of the head posture angle be α0, and the angle change to α during the reading action. t By calculating the angle difference Δα=α t -Δ0, when Δα is within the preset reading angle range, the head-down reading action is considered valid;

[0017] As a further improvement of the present invention, in step S3, the single tile inspection process is repeated 3 times.

[0018] As a further improvement of the present invention, in step S3, the number of times the balloon is squeezed by hand is 5-6 times.

[0019] As a further improvement of the present invention, an alarm is triggered for non-compliant operations in steps S2 and S3 to promptly remind monitoring personnel and gas inspectors.

[0020] As a further improvement of the present invention, the alarm includes visual signals, sound signals, and electronic notifications.

[0021] The beneficial effects of this invention are:

[0022] 1. The tile inspection operation compliance identification method based on target tracking algorithm constructed in this invention has shown significant results in improving the standardization and supervision of tile inspection operations. By integrating target identification and target tracking, it establishes strict standards and verification processes for each link of tile inspection operations.

[0023] (1) In the target recognition stage, with the help of algorithms such as YOLO based on convolutional neural networks, the tile inspector and the tile inspection device he carries can be accurately located.

[0024] (2) Target tracking technology ensures accurate monitoring of key actions of tile inspectors. For example, when holding the tile detector high, the accuracy of its position is judged by setting the detection area range and combining coordinates; when squeezing a balloon, the algorithm monitors the changes in the position of the hand joints and judges whether it meets the requirements by calculating Euclidean distance; when looking down to read the value, the accuracy of the action is confirmed by tracking the changes in the head posture angle.

[0025] 2. This invention can effectively reduce human error and operational deviations. In traditional tile inspection operations, human factors can easily lead to various violations, such as incorrect positioning of the tile detector, incorrect number of times the balloon is squeezed, and incorrect data recording. The automated identification and verification mechanism of this method can monitor the operation steps in real time, correct errors and deviations in a timely manner, and ensure strict adherence to safety standards.

[0026] 3. Real-time video monitoring and automatic data recording greatly enhance the transparency of operations. High-definition video cameras record the entire operation process in real time under various environments, allowing management to intuitively understand the work of gas inspectors; the automatic data recording function not only records video information but also extracts and stores relevant operation data, facilitating management review and analysis, helping to identify problems in a timely manner and optimize work processes.

[0027] 4. This invention can effectively prevent safety hazards such as gas explosions. By monitoring gas inspectors' operations in real time, it ensures the accuracy and timeliness of gas detection, avoiding undetected gas accumulation due to missed or false detections. This measure protects the safety of workers and reduces economic losses caused by accidents.

[0028] 5. The real-time alarm mechanism enhances emergency response capabilities. Once the system detects non-compliant operations, such as gas inspectors repeating operations without following regulations, incomplete operations, or incorrect data recording, an alarm is immediately triggered, notifying regulatory personnel and gas inspectors to take timely corrective measures and prevent accidents from occurring. Attached Figure Description

[0029] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] Figure 1 This is a flowchart illustrating the operation of the compliance identification method for tile inspection operations based on the target tracking algorithm of the present invention.

[0031] Figure 2 This is a flowchart of the compliance judgment criteria for gas inspection.

[0032] Figure 3 This is a feature map of the target tracking algorithm for tile inspection operations;

[0033] Figure 4 This is a diagram of the target tracking algorithm model architecture for tile inspection operations. Detailed Implementation

[0034] Please refer to Figure 1 The present invention provides a compliance identification method for tile inspection operations based on a target tracking algorithm, comprising the following steps:

[0035] Step S1: Deploy a high-definition video camera in the methane detection area of ​​the mine to collect video data in the methane detection area in real time and preprocess the images.

[0036] Methane detection zones are selected at key locations within the mine, such as areas prone to gas accumulation, along gas inspectors' regular work routes, and near important equipment. High-definition video cameras acquire real-time images of the methane detection zones.

[0037] 1. Due to the low light levels in the mine, this embodiment places high demands on the high-definition video camera to obtain high-quality images. The camera should have wide dynamic range (WDR) functionality. Let the light intensity range in the mine be I. min To I max The dynamic range (DR) of the selected camera should meet the following requirements. Configure the camera parameters, including resolution settings such as 1920×1080; frame rate settings such as 25 frames per second; focus and exposure settings, to ensure that clear images can be captured for subsequent analysis.

[0038] 2. Preprocess the real-time acquired video images to provide more accurate input data for subsequent motion recognition.

[0039] Specifically, video data is acquired in real time, and the acquired video stream is denoted as V(t), where t represents time.

[0040] (1) Image denoising

[0041] Using the mean filtering denoising method, for each pixel (x, y) in the image, its neighborhood size is set to n×n (e.g., 3×3). Then the denoised pixel value P... denoise The formula for calculating (x,y) is:

[0042]

[0043] Where P(x+i,y+j) is the original value of the pixel in the neighborhood.

[0044] (2) Image enhancement processing

[0045] Histogram equalization is used to enhance image contrast. Let the original image have a grayscale range of [0, L-1], a total of N pixels, and the frequency of grayscale level k be nk. Then its probability density function... After histogram equalization, the formula for calculating the new gray level T(k) corresponding to the gray level is:

[0046] Step S2: Using the YOLO target detection method, identify the gas inspector and gas detector in the methane detection area in the video; if the gas inspector is detected holding a gas detector and arrives at the designated methane detection area on time, it is determined to be a compliant operation; otherwise, it is determined to be a non-compliant operation.

[0047] To ensure that gas inspectors arrive at their designated work locations on time and carry their gas inspection equipment, this system employs target detection image recognition technology. This technology focuses on identifying specific features in images—namely, the behavior of gas inspectors carrying their gas inspection equipment.

[0048] This algorithm takes "the gas inspector holding a gas detector" as a key feature. This feature involves identifying the posture of the person and the object in their hand - the gas detector. In this way, the target object can be accurately identified from multiple similar image features. Let the person posture feature vector be The appearance feature vector of the gas detector is Then the combined feature vector is used to identify the gas inspector and the gas detector he carries.

[0049] This embodiment uses the convolutional neural network YOLO algorithm to implement object detection. This algorithm can quickly and accurately locate the people and gas detectors in the image when viewing the image once. Taking an image with a size of 480×640×3 from the gas inspection operation site as input, this image contains rough roadway wall textures, lighting equipment with different brightness and colors, as well as key target objects such as gas inspectors and gas detectors. The convolutional neural network YOLO algorithm implements object detection based on a unique grid division strategy, and the specific steps are as follows:

[0050] (1) Image grid division: The input image is evenly divided into S×S grids, where S = 13. Each grid cell (i, j) (where 0 ≤ i < S, 0 ≤ j < S) is responsible for detecting the target objects falling into this grid. This division method discretizes the entire image space, enabling the algorithm to search and locate targets in different local areas.

[0051] (2) Grid cell prediction:

[0052] Each grid cell (i, j) is responsible for predicting B bounding boxes, where B = 2, as well as the confidence of these bounding boxes and C class probabilities. In the gas inspection operation scenario, C = 2.

[0053] For each grid cell (i, j), its prediction result is obtained through the calculation of the convolutional neural network. The size of the convolutional kernel is 3×3, and the convolutional operation at the grid cell (i, j) can be expressed by the following formula:

[0054]

[0055] Where is the output feature value of the l-th convolutional kernel at the grid cell (i, j), is the weight of the l-th convolutional kernel at the position (m, n), b l is the corresponding bias term, I(i + m, j + n) represents the pixel value of the input image at the position (i + m, j + n), and * represents the convolutional operation.

[0056] The prediction result of the grid cell (i, j) can be expressed as a tensor T ij , where p cThis represents the probability of category c, where c=1 represents a tile inspector and c=2 represents a tile detector. (x,y,w,h) are the coordinates and dimensions of the bounding box. Here, (x,y) is the offset of the bounding box center relative to the grid cell, ranging from (0,1). w and h are the proportions relative to the entire image size. The calculation method is as follows:

[0057]

[0058] Where f ij It is the eigenvalue at grid cell (i,j), w ck ,w xk ,w yk ,w wk ,w hk These are the corresponding convolutional kernel weights, b c ,b x ,b yj ,b w ,b hz σ is the bias term, σ is the activation function (such as the sigmaoid function), and e is the natural exponential function, used to calculate the width and height of the bounding box.

[0059] Finally, YOLO outputs an S×S×(B×5+C) tensor, representing the prediction result for each grid cell in the entire image. In this embodiment, with a 13×13 grid division, the output tensor size is 13×13×(2×5+2). By parsing this tensor, the bounding box information and class probability of each grid cell can be obtained. A confidence threshold of 0.5 is set to filter reliable detection results, and non-maximum suppression is performed to remove bounding boxes with high overlap, thus obtaining the final accurate detection location and class information for the tile inspector and the tile inspection device. This information will provide crucial basis for subsequent compliance judgments of tile inspection operations, helping the system accurately determine whether the tile inspector arrives at the designated location on time and whether they carry the necessary tile inspection equipment, thereby effectively improving the safety and standardization of the entire tile inspection operation.

[0060] When the detected target feature vector matches the predefined feature vector of "tile inspector holding tile inspection device" to a threshold θ, which is set to 0.8, the presence of the tile inspector is confirmed.

[0061] Step S3: After verifying that the gas inspector arrives at the designated methane detection area on time with the gas detector, the CSRT target tracking algorithm is used to track the gas detection operation in real time. Then, the operation is compared with standardized procedures. If the gas detection operation conforms to the standardized procedures, it is considered a compliant operation; otherwise, it is considered a non-compliant operation. To ensure that gas inspectors follow the prescribed operating procedures during gas detection, this system uses an advanced target tracking algorithm to identify and verify key actions.

[0062] The technical implementation process of this step is explained in detail below. Please refer to [link / reference]. Figures 2-4 .

[0063] 1. Action definition and target identification:

[0064] The system defines three key actions: the gas inspector raises the gas detector, squeezes the gas balloon, and lowers their head to read the reading. These actions are considered standardized operating procedures for gas detection operations and need to be executed and identified accurately.

[0065] 2. Deployment of target tracking algorithms:

[0066] The Channel and Spatial Reliability Tracker (CSRT) algorithm enables real-time tracking of inspectors' movements within a video stream, maintaining continuous visual continuity of the target. The CSRT algorithm searches for the region most similar to the target within video frames by calculating the color histogram and spatial information of the target area.

[0067] (1) Determining the Initial Target Region: In the actual scenario of tile inspection operations, the initial position of the target is determined by first detecting the tile inspector or tile detector in the starting frame of the video stream using a target detection algorithm, such as SSD or YOLO. Assume the position and size of the detected tile inspector in the first frame are represented by (x0, y0, w0, h0) (where x0, y0 are the coordinates of the upper left corner of the target region, and w0, h0 are the width and height of the target region), and the initial target region R0 is defined accordingly. For example, in a video frame with a resolution of 480×640, the initial position of the tile inspector might be (100, 200, 50, 150), that is, an area starting from coordinates (100, 200), with a width of 50 pixels and a height of 150 pixels.

[0068] (2) Color Space Quantization and Histogram Calculation: The pixels within the initial target region R0 are quantized using the common RGB color space, divided into 8×8×8 intervals. For each pixel within region R0, its quantization interval is determined based on its RGB value. Then, the number of pixels in each interval is counted to obtain the color histogram H0 of the target region R0. This color histogram H0 will serve as the color feature representation of the target, used for subsequent searching of regions with similar colors to the target in video frames.

[0069] (3) Spatial information extraction: In addition to color features, spatial information of the initial region of the target is extracted, including its position (x0, y0) and size (w0, h0). This spatial information plays an important role in the target tracking process and, together with color features, is used to accurately identify changes in the target's position in subsequent frames.

[0070] (4) Video stream data acquisition: Continuously acquire video streams of the gas inspection operation scene collected by high-definition video cameras deployed in the mine. Each frame of the video stream will be used as input data for target tracking. Its image size is consistent with the resolution configured by the camera, such as 480×640×3 (where 3 represents the three color channels of RGB), and the frame rate, such as 25 frames / second, determines the temporal resolution of target tracking, that is, there are 25 opportunities per second to update the target's position information.

[0071] 3. Algorithm

[0072] (1) Objective Function Construction and Significance: The CSRT algorithm searches for the region most similar to the target in video frames by calculating the color histogram and spatial information of the target region. Let R be the position of the target in frame t. t =(x t ,y t ,w t ,h t (Position coordinates and dimensions), by minimizing the objective function To update the target position. Where p i It is the candidate region R t Features of pixel i (including color histogram and spatial information), p t These are the target features (i.e., the color histogram H0 and spatial information (x0, y0, w0, h0) of the initial target region R0), w i ρ is the weight, ρ is the distance metric (such as Bhattacharyya distance), and λ is the regularization parameter.

[0073] (2) Feature extraction and representation: For candidate region R t For each pixel i within the range, its feature p i This includes color histogram features and spatial information features. The color histogram features are obtained using the same quantization method as calculating the color histogram of the initial target region R0. The spatial information features include the position of pixel i within the candidate region R0. t The relative position coordinates within the region. For example, for candidate region R... t For a pixel i within a given range, its color histogram feature might be an 8×8×8 vector representing the distribution of that pixel's color across various quantization intervals. Its spatial information feature might be a two-dimensional vector (x...). ir el,y ir el), indicating that the pixel is relative to the candidate region R. t The coordinate offset of the top left corner.

[0074] (3) Distance metric calculation: The distance metric function ρ is used to measure the candidate region R. t Features of inner pixel i and target feature p iThe differences between them. Taking the Bach distance as an example, for the color histogram H... i (Color histogram of candidate regions) and H t (Color histogram of the target area), the formula for calculating the Bach distance is: Where K is the number of quantization intervals in the color histogram, here K = 8 × 8 × 8). For spatial distance calculations, Euclidean distance can be used, for example, for location coordinates (x... irel ,y irel (x0, y0) are the relative position coordinates of pixels within the candidate region and (x0, y0) are the initial position coordinates of the target region, with a spatial distance of . ρ([p) is obtained by combining color and spatial distance. i -p t ] 2 ).

[0075] (4) Weight Calculation and Optimization: Weights play a role in adjusting the contribution of different pixels to the target location estimation in the algorithm. Initially, the weights can be initialized based on the pixel position distribution within the target region, with larger weights for pixels in the central region and smaller weights for pixels in the edge region. Then, during the objective function optimization process, the weights w are continuously adjusted by minimizing the objective function J(t). i The value of is determined by the similarity between the current candidate region and the target region. In each iteration, the weight of pixels with high similarity is increased, and the weight of pixels with low similarity is decreased, based on the similarity between the current candidate region and the target region, thus making the estimation of the target location more accurate.

[0076] (5) Regularization: The regularization parameter λ is used to prevent overfitting and ensure the stability of the objective function during the optimization process. It affects the weights w i norm ||w|| 2 Constraints are applied to prevent excessive weights from making the model overly complex and losing its generalization ability. The value of the regularization parameter λ is usually determined experimentally or empirically. In target tracking scenarios during inspection operations, adjustments may be made based on the characteristics of the actual data and the tracking effect. When there are significant changes in lighting or frequent changes in the target's appearance, the value of λ should be appropriately adjusted to balance the model's fitting ability and stability.

[0077] (6) Target position update iteration: In each frame of video image, candidate regions are moved within the search area (the size of the search area can be dynamically adjusted according to the target's movement speed and scene complexity, initially set to 2-3 times the size of the target's initial region R0), and the objective function value J(t) of each candidate region is calculated. Then, the candidate region position that minimizes the objective function value is selected as the new position estimate R of the target in the current frame. t =(x t ,y t ,w t ,ht Next, based on the new target location R, update the target feature p. t (Including color histogram and spatial information), and repeat the above process to iteratively optimize the target location estimation until the objective function value converges or the preset number of iterations is reached.

[0078] 4. Output

[0079] (1) Target position update result: After the above complex calculation and iteration process, the updated position R of the target---gas inspector or gas detector in each frame is finally obtained. t =(x t ,y t ,w t ,h t This location information will serve as a crucial input for subsequent action recognition and verification steps, determining whether the tile inspector performed actions such as raising the tile detector, squeezing the balloon, and looking down to read the value, and whether these actions were performed within the designated area. If it is stipulated that the tile detector should be located within a specific area when raised, the location of the tile detector (x) will be tracked. t ,y t ,w t ,h t This allows for comparison with the area to determine whether the action is compliant.

[0080] (2) Guarantee of continuous visual continuity: By repeating the target tracking process described above in each frame of the video stream, the CSRT algorithm can maintain continuous visual continuity of the target. This means that the system can monitor the target's motion trajectory in real time and accurately track the behavioral changes of the gas inspector throughout the entire gas inspection operation. Whether the gas inspector is walking in the tunnel, performing inspection operations, or interacting with other equipment or personnel, the system can continuously track its position, providing a reliable basis for comprehensively assessing the compliance of the gas inspection operation and ensuring that the judgment of compliance will not be wrong due to the interruption or inaccuracy of target tracking.

[0081] 5. Action Recognition and Verification

[0082] The action recognition and verification process mainly consists of three steps: raising the detector, squeezing the balloon, and lowering the head to read the value. The specific steps are as follows.

[0083] (1) High-lift methane detector: The system is set with a specific methane detection range of R = (x min ,y min ,x max ,y max When the detector is raised to the range (x, y), x satisfies min ≤x≤x max And y min ≤y≤ymax At that time, the system confirms the execution of the action through target tracking.

[0084] (2) Squeezing the balloon by hand: The system uses a target tracking algorithm to continue monitoring the inspector's hand movements to ensure that the balloon is correctly and continuously squeezed a specified number of times.

[0085] Let the initial coordinates of the key points of the hand be... The position coordinates in subsequent frames are By calculating the Euclidean distance When d t Within the preset pinching distance range (e.g., [d] min ,d max If the action of squeezing the balloon is deemed valid when the number of consecutive times reaches a preset value N (e.g., 5 times), the action is considered valid.

[0086] (3) Head down reading: When the operation of the tile detector enters the reading stage, the system tracks the head movement of the tile inspector to confirm whether the reading step has been performed correctly.

[0087] Let the initial value of the head posture angle be α0, and the angle change to α during the reading action. t By calculating the angle difference Δα = α t -α0, when Δα is within the preset reading angle range (e.g., [α]). min ,α max When the reading action is taken, it is determined that the act of looking down is valid.

[0088] 6. Identification of non-compliance in gas inspection operations:

[0089] Establish compliance criteria for tile inspection operations. If a tile inspector's inspection process deviates from these criteria, the inspection is deemed non-compliant. Specific criteria are as follows: Figure 2 As shown.

[0090] (1) Determine whether the gas inspector arrived on time.

[0091] The arrival time T of the gas inspector was analyzed through video images. arrive With the specified arrival time T required When T arrive >T required Furthermore, if a tile inspection device is not carried, the tile inspection operation is deemed non-compliant.

[0092] (2) Determine whether the gas inspector's actions are compliant.

[0093] The actions of gas inspectors are assessed using video footage to determine if they meet requirements. The procedure for a single methane concentration test is divided into three steps: the inspector raises the gas detector, the inspector squeezes the gas balloon, and the inspector lowers their head to read the value. The inspector must repeat this procedure three times. For each of the three steps in a single methane concentration test, the following situations are considered non-compliant:

[0094] ① Raising the methane detector: If the number of times the methane detector fails to reach the methane detection area exceeds the threshold N1 (e.g., once) during a single detection process, this action is deemed non-compliant.

[0095] ② Squeezing balloons: If the number of times the balloon is squeezed does not reach the specified number N2, such as 5 times, or the action is not continuous, the action is judged to be non-compliant.

[0096] ③ Head down reading: If no head down reading action is detected or the head posture angle is not within the specified range, this action is deemed non-compliant.

[0097] If the inspector does not repeat the above procedure three times, it will also be deemed non-compliant.

[0098] This method also includes an alarm triggering mechanism: when the system detects the above-mentioned non-compliance, such as the gas inspector not operating according to regulations, incomplete operation, or incorrect data recording, an alarm will be triggered immediately, and the on-site gas inspector will be notified to correct the action through flashing lights, voice alarms, and system notifications. At the same time, the non-compliance information will be sent to relevant management personnel for processing.

[0099] Gas inspection operations are conducted underground in mines. This invention first installs high-definition video cameras at key locations in the mine to ensure clear images and high recognition rates. Then, it uses the YOLO algorithm, a neural network model, to identify the gas inspector and their handheld gas detector, determining whether they have arrived at the designated site on time. After confirming arrival, the CSRT target tracking algorithm tracks three prescribed actions in real time, comparing them with standardized gas inspection procedures to determine if they have been performed and if the required number of actions have been taken. After completing the three actions, the gas inspection information is recorded on the site log. For non-compliant gas inspection operations, an alarm is triggered to correct the non-compliant behavior. This technical solution effectively avoids false inspections, missed inspections, and incorrect inspections in gas inspection operations, thereby ensuring the compliance and reliability of gas inspection work.

[0100] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes that can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention are all within the protection scope of the claims of the present invention.

Claims

1. A method for compliance identification of tile inspection operations based on a target tracking algorithm, characterized in that, Includes the following steps: Step S1: Deploy a high-definition video camera in the methane detection area of ​​the mine to collect video data in the methane detection area in real time and preprocess the images. Step S2: Using the YOLO target detection algorithm, identify the gas inspector and gas detector in the methane detection area in the video; if the gas inspector is detected holding a gas detector and arrives at the designated methane detection area on time, it is determined to be a compliant operation; otherwise, it is determined to be a non-compliant operation. Step S3: After verifying that the gas inspector arrives at the designated methane detection area on time with the gas detector in hand, the gas inspection operation on site is tracked in real time using the CSRT target tracking algorithm; then it is compared with the set standardized operation. If the gas inspection operation conforms to the standardized operation, it is judged as a compliant operation; otherwise, it is judged as a non-compliant operation. Standardized operations include: (1) Single gas detection procedure: Hold the gas detector high, squeeze the balloon and lower your head to read the value; (2) Repeat the single detection action process according to the set number of times.

2. The compliance identification method for tile inspection operations based on target tracking algorithm according to claim 1, characterized in that, In step S1, the high-definition video camera has a wide dynamic range recognition function, and the light intensity range inside the mine is set to I. min To I max The selected camera's wide dynamic range (DR) meets the requirements.

3. The compliance identification method for tile inspection operations based on target tracking algorithm according to claim 1, characterized in that, In step S2, the method for determining whether the gas inspector arrived on time is to analyze the arrival time T of the gas inspector through video image analysis. arrive With the specified arrival time T required When T arrive >T required At that time, it was determined that the inspection operation was not in compliance with regulations.

4. The compliance identification method for tile inspection operations based on target tracking algorithm according to claim 1, characterized in that, In step S3, the method for comparing on-site tile inspection operations with standardized operations is as follows: (1) Action determination of the high-lift gas detector: Let the methane detection area be R = (x min ,y min ,x max ,y max When the gas detector is raised to the range (x, y) and x satisfies min ≤x≤x max and y min ≤y≤y max At that time, the execution of the action is confirmed through target tracking; (2) Determining the action of squeezing the balloon: A target tracking algorithm is used to monitor the hand movements of the gas inspector. The initial coordinates of the key points on the hand are set as the coordinates of the inspector's hand in the first frame, denoted as... The position coordinates in subsequent frames are By calculating the Euclidean distance When d t The action of squeezing the balloon by hand is considered valid when the number of times the squeezing distance is within the preset range and the preset value N is met consecutively. (3) Determining the action of looking down to read: Let the initial value of the head posture angle be α0, and the angle change to α during the reading action. t By calculating the angle difference Δα=α t -α0, when Δα is within the preset reading angle range, the head-down reading action is considered valid.

5. The compliance identification method for tile inspection operations based on target tracking algorithm according to claim 1, characterized in that, In step S3, the single tile inspection process is repeated 3 times.

6. The compliance identification method for tile inspection operations based on target tracking algorithm according to claim 1, characterized in that, In step S3, the balloon is squeezed 5-6 times consecutively.

7. The compliance identification method for tile inspection operations based on the target tracking algorithm according to claim 1, characterized in that, An alarm will be triggered for any non-compliant operations in steps S2 and S3, promptly alerting monitoring personnel and gas inspectors.

8. The compliance identification method for tile inspection operations based on the target tracking algorithm according to claim 7, characterized in that, The alarm includes visual signals, sound signals, and electronic notifications.

Images