Detection method for safe operation of vehicles in coal mine transport tunnels

Abstract

This invention discloses a method for detecting the safe operation of a coal mine transport vehicle. The method includes: collecting field data to train a target detection model and setting scene detection parameters; acquiring images and processing them using the target detection model to obtain position information of the vehicle and pedestrians; analyzing the vehicle position information using a vehicle tracking unit to determine the operating status of the vehicle; analyzing the position of pedestrians using a personnel monitoring unit to obtain information about people in hazardous areas; and combining the operating status of the vehicle and the information about people in hazardous areas to determine whether there are current safety hazards. This invention can address the issue of monitoring the standardized use of vehicle in coal mine mining projects and achieve intelligent monitoring of the safe operation of vehicle in coal mines.

Description

Technical Field

[0001] The invention belongs to the technical field of coal mine safety detection, and in particular relates to a method for detecting safe operation of vehicles in coal mine transport tunnels. Background Art

[0002] Cross-country man-lifts (CMMs) are widely used in coal mining operations, reducing the burden on personnel and increasing mobility. They are a common means of transporting people. Unlike cars, CMMs are construction vehicles, making them less safe than ordinary vehicles and exhibiting poor braking performance. Therefore, their operation requires enhanced safety oversight. However, irregular boarding and alighting procedures are common. Coal miners, lacking safety awareness, often disembark prematurely to save a few seconds or board after the vehicle has already started. Disembarking before a CMM has come to a complete stop or boarding after the vehicle has already started is extremely dangerous. If a person falls while boarding or exiting, the CMM's poor braking performance makes it difficult to stop the vehicle quickly, potentially leading to an accident. Therefore, the standardized use of CMMs is crucial for ensuring the safety of personnel in coal mines.

[0003] Traditionally, safety officers have been used to regulate personnel behavior to prevent accidents caused by improper use. However, this approach is time-consuming and subject to significant human influence. Furthermore, labor costs are high, which impacts transportation efficiency and fails to meet safety management requirements. Summary of the Invention

[0004] In order to solve the labor cost and efficiency problems of traditional methods, the present invention discloses a method for detecting the safe operation of vehicles in coal mine transportation tunnels. The position and motion state of the vehicle are detected by a deep learning-based target detection algorithm, and real-time behavioral safety monitoring is achieved by combining it with the position of people in the environment.

[0005] The present invention discloses a method for detecting safe operation of a manned vehicle in a coal mine transport tunnel, comprising the following steps:

[0006] Step S1: Collect data from the monitoring site to train the target detection model, delineate polygonal electronic fences according to specific circumstances, and set scene detection parameters;

[0007] Step S2: Obtain images captured by a camera placed near a parking spot for vehicles in a lane, and use a target detection model to identify and process the images to obtain location information of the vehicle compartment and pedestrians;

[0008] Step S3: Analyze the carriage position information through the carriage tracking unit to obtain the running status of the level tunnel man-car;

[0009] Step S4: Analyze the position of pedestrians through the personnel monitoring unit to obtain information about people in the dangerous area;

[0010] Step S5: When a pedestrian is detected in a dangerous area while the tunnel vehicle is in operation, it is determined that there is a safety hazard in the tunnel vehicle, an alarm is triggered, and an image is captured at the time of the alarm, the position of the pedestrian is marked and saved for inspection.

[0011] Furthermore, the step S1 specifically includes:

[0012] Step S1.1: Select a camera position that can clearly capture the boarding and alighting locations and deploy image acquisition equipment;

[0013] Step S1.2: Simulate the scene of people and vehicles running in the tunnel and collect sample images;

[0014] Step S1.3: Label the car and pedestrian objects in the image and export the data required for training the object detection model;

[0015] Step S1.4: Using the target detection network as the basic framework, transfer learning is performed on the training data to obtain the boarding and alighting detection model.

[0016] Step S1.5: Set the car confidence threshold C according to the model detection effect car and pedestrian confidence threshold C person ,Detection results below the set confidence threshold will be discarded;

[0017] Step S1.6: Based on the collected images, the track and vehicle travel range are delineated to set a polygonal danger zone, and the absolute position sequence of the vertices forming the polygon in the image is recorded;

[0018] Step S1.7: According to the scene conditions, set the maximum pixel value MAX_TOLERANCE between the two frames of the carriage movement. If the detected carriage position changes exceed this value, it is considered to be two different carriages;

[0019] Step S1.8: Set the number of retained frames N. The vehicle position tracking will refer to the image data of the most recent N frames for analysis;

[0020] Step S1.9: Set the minimum moving distance M based on the model detection effect of the carriage min ,The carriage position change below this value is regarded as a detection error;

[0021] Step S1.10: Set the pedestrian object determination anchor point (A xs , A xe , A ys , A ye ) is used as the criterion for determining whether a pedestrian has entered a dangerous area, where Axs is the starting percentage of horizontal interception, A xe is the end percentage of horizontal interception, A ys is the starting percentage of vertical interception, A ye is the end percentage of the longitudinal intercept;

[0022] Step S1.11: Set the intrusion ratio threshold T a ,When the portion of the pedestrian object intercepted by the anchor point that intrudes into the dangerous area accounts for a proportion of its own area that exceeds this value, the object is considered to be in the dangerous area;

[0023] Furthermore, step S3 specifically includes:

[0024] Step S3.1: Maintain a carriage position record table, which stores the position information of each tracked carriage in the latest N frames of images Where N is the number of retained frames, is the position information quadruple of the carriage that appears n times before the last one tracked in section i. The position information quadruple consists of four values (x1, y1, x2, y2), where x1 and y1 are the horizontal and vertical coordinates of the upper left corner of the carriage detection box, and x2 and y2 are the horizontal and vertical coordinates of the lower right corner of the carriage detection box, respectively.

[0025] Step S3.2: Filter out the compartments whose confidence level is not less than the confidence threshold C. car The carriage inspection results;

[0026] Step S3.3: Take out one of the carriage detection results and calculate the distance D = {d1, d2, ..., d j}, where j is the number of carriages in the current carriage position record table, and the carriage distance is the Euclidean distance. and the position information quadruple of carriage b The method to calculate the distance d between them is:

[0027]

[0028] Find the minimum distance d min ∈D, if d min If the value is not less than the tolerance range MAX_TOLERANCE, the carriage detection result is placed in the carriage tracking record corresponding to the minimum distance; otherwise, the carriage detection result is added to the carriage position record table as a new carriage tracking record.

[0029] Step S3.4: Repeat step S3.3 until all the filtered carriage detection results are updated;

[0030] Step S3.5: Check each carriage tracked in the carriage position record table. If there is any carriage whose total movement distance in the latest N retained frames exceeds the minimum movement distance M min It is determined that the level tunnel manned vehicle is running.

[0031] Furthermore, step S4 specifically includes:

[0032] Step S4.1: Filter out pedestrians whose confidence level is not less than the pedestrian confidence threshold C person Pedestrian detection results;

[0033] Step S4.2: Determine the anchor point (A) based on the pedestrian object xs , A xe , A ys , A ye ) Intercept the pedestrian detection frame to obtain the pedestrian object judgment frame, and calculate the area of the pedestrian object judgment frame as Area p ;

[0034] Step S4.3: Use the polygonal electronic fence to intersect with the pedestrian object judgment box, calculate the area of the intersection part and obtain the area of the pedestrian object judgment box intruding into the danger zone as Area i ; Calculate the ratio of the area of the pedestrian object judgment box invading the dangerous area to the area of the pedestrian object judgment box itself R = Area i ÷Area p ;

[0035] Step S4.4: If the ratio R of the area of the pedestrian object judgment frame intruding into the dangerous area to the area of the pedestrian object judgment frame itself is greater than or equal to the intrusion ratio threshold T a It is judged that the pedestrian is in a dangerous area.

[0036] The advantages and positive effects of the present invention are:

[0037] 1. The present invention has strong expansion capabilities and can be used with any target detection algorithm. It has flexible deployment and can meet the flexible requirements of different scenarios for equipment performance. The deployment equipment can be flexibly selected based on the actual scenario requirements, which greatly reduces the material cost of system deployment.

[0038] 2. The present invention does not require the object detection model to be able to detect all objects 100% correctly. A small number of falsely detected and missed objects can still be ignored by the tracking strategy without affecting the final discrimination effect. Therefore, the present invention has strong robustness.

[0039] 3. The present invention has strong scalability, and the detection method of the present invention can be applied to similar security supervision scenarios by making simple modifications.

[0040] 4. The present invention effectively solves the problem of operational safety supervision of vehicles in horizontal tunnels in mining projects, and can significantly reduce the cost of personnel transportation safety assurance in coal mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart for analyzing on-site images using a coal mine transport lane safety vehicle operation detection method;

[0042] Figure 2 This is a schematic diagram of using pedestrian object determination anchor points to capture images;

[0043] Figure 3 It is a flow chart for updating carriage detection results;

[0044] Figure 4 It is a flowchart for updating pedestrian detection results;

[0045] Figure 5 The figure shows the detection effect of a stopped lane vehicle in a mine according to an embodiment of the present invention;

[0046] Figure 6 This is a diagram illustrating the detection effect of an embodiment of the present invention on a tunnel-mounted vehicle leaving a station in a mine;

[0047] Figure 7 This is an illustration of the detection effect of an embodiment of the present invention on a tunnel-entering vehicle in a mine, where a person is illegally getting off the vehicle. DETAILED DESCRIPTION

[0048] The present invention will be further described in detail below with reference to specific embodiments and drawings. The following embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operating processes, but the scope of protection of the present invention is not limited to the following embodiments.

[0049] The embodiment of the present invention provides a method for detecting safe operation of vehicles in coal mine transport tunnels. The main process is as follows: Figure 1 As shown, the specific steps include:

[0050] Step S1: Collect data from the monitoring site to train the target detection model, delineate polygonal electronic fences according to specific circumstances, and set scene detection parameters.

[0051] Specifically, the vehicle transportation personnel and the safety department are coordinated to simulate some dangerous behaviors on the basis of ensuring personnel safety, such as getting off the vehicle before it stops and staying in a dangerous area after the vehicle has started. Video clips of the above dangerous behaviors are collected through cameras. Frame images of the video are extracted and all pedestrians in the picture are marked. Easy-to-distinguish car features are selected and marked. Training data is derived from the above-mentioned marked images, and the target detection network is used as the basic framework to perform transfer learning on the training data to obtain an on-and-off vehicle detection model.

[0052] According to the model detection effect, the compartment confidence threshold C is set. car and pedestrian confidence threshold C person , detection results below the set confidence threshold will be discarded; according to the scene image shooting angle, the track and vehicle travel range are delineated to set a polygonal danger zone, and the absolute position sequence of the vertices forming the polygon in the image is recorded; the maximum pixel value MAX_TOLERANCE of the movement between two frames of the carriage is set, and the detected carriage position change exceeding this value is regarded as two different carriages; the number of retained frames N is set, and the carriage position tracking will refer to the image data of the latest N frames for analysis; set the minimum movement distance M min , the change of the carriage position below this value is regarded as the detection error; set the pedestrian object determination anchor point (A xs , A xe , A ys , A ye ) is used as the criterion for determining whether a pedestrian has entered a dangerous area, where A xs is the starting percentage of horizontal interception, A xe is the end percentage of horizontal interception, A ys is the starting percentage of vertical interception, A ye is the vertical interception end percentage, Figure 2 Schematic diagram of using pedestrian object anchor point to intercept the image; setting the intrusion ratio threshold T a ,When the portion of the pedestrian object intercepted by the anchor point that intrudes into the dangerous area accounts for a proportion of its own area that exceeds this value, the object is considered to be in the dangerous area;

[0053] Step S2: Obtain images captured by a camera placed near a parking spot for vehicles in a lane, and use a target detection model to identify and process the images to obtain location information of the vehicle compartment and pedestrians;

[0054] Step S3: Analyze the carriage position information through the carriage tracking unit to obtain the running status of the level tunnel man-car;

[0055] Specifically, step S3 includes the following steps:

[0056] Step S3.1: Maintain a carriage position record table, which stores the position information of each tracked carriage in the latest N frames of images Where N is the number of retained frames, is the position information quadruple of the carriage that appears n times before the last one tracked in section i. The position information quadruple consists of four values (x1, y1, x2, y2), where x1 and y1 are the horizontal and vertical coordinates of the upper left corner of the carriage detection box, and x2 and y2 are the horizontal and vertical coordinates of the lower right corner of the carriage detection box, respectively.

[0057] Step S3.2: Filter out the compartments whose confidence level is not less than the confidence threshold C. car The carriage inspection results;

[0058] Step S3.3: Take out one of the carriage detection results and calculate the distance D = {d1, d2, ..., d j}, where j is the number of carriages in the current carriage position record table; find the minimum distance d min ∈D, if d min If the value is not less than the tolerance range MAX_TOLERANCE, the carriage detection result is placed in the carriage tracking record corresponding to the minimum distance; otherwise, the carriage detection result is added to the carriage position record table as a new carriage tracking record. Figure 3 This is a flow chart for updating carriage detection results.

[0059] Step S3.4: Repeat step S3.3 until all the filtered carriage detection results are updated;

[0060] Step S3.5: Check each carriage tracked in the carriage position record table. If there is any carriage whose total movement distance in the latest N retained frames exceeds the minimum movement distance M min It is determined that the level tunnel manned vehicle is running.

[0061] Step S4: Analyze the position of pedestrians through the personnel monitoring unit to obtain information about people in the dangerous area;

[0062] Specifically, step S4 includes the following steps:

[0063] Step S4.1: Filter out pedestrians whose confidence level is not less than the pedestrian confidence threshold C person Pedestrian detection results;

[0064] Step S4.2: Determine the anchor point (A) based on the pedestrian object xs , A xe , A ys , A ye) Intercept the pedestrian detection frame to obtain the pedestrian object judgment frame, and calculate the area of the pedestrian object judgment frame as Area p ;

[0065] Step S4.3: Use the polygonal electronic fence to intersect with the pedestrian object judgment box, calculate the area of the intersection part and obtain the area of the pedestrian object judgment box intruding into the danger zone as Area i ; Calculate the ratio of the area of the pedestrian object judgment box invading the dangerous area to the area of the pedestrian object judgment box itself R = Area i ÷Area p ;

[0066] Step S4.4: If the ratio R of the area of the pedestrian object judgment frame intruding into the dangerous area to the area of the pedestrian object judgment frame itself is greater than or equal to the intrusion ratio threshold T a It is judged that the pedestrian is in a dangerous area.

[0067] Figure 4 This is the pedestrian detection result update flow chart.

[0068] Step S5: When a pedestrian is detected in a dangerous area while the tunnel vehicle is in operation, it is determined that there is a safety hazard in the tunnel vehicle, an alarm is triggered, and an image is captured at the time of the alarm, the position of the pedestrian is marked and saved for inspection.

[0069] The effect of the present invention can be further verified by the performance of this embodiment in actual scenarios.

[0070] Figure 5 This is detection software implemented using the steps and methods of this embodiment. The software is loaded with the target detection model and scene detection parameters trained in step S1. The figure shows the software detecting a stopped vehicle in a mine tunnel. It can be seen that the vehicle window is used as a discriminant feature in this embodiment. The software uses the vehicle tracking unit described in step S3 of this method to determine that the vehicle is stopped and not currently engaging in dangerous behavior.

[0071] Figure 6 This is the detection effect of using detection software to detect a tunnel-mounted vehicle leaving a station in a mine. It can be seen that the vehicle tracking unit analyzes the motion trajectory of the tunnel-mounted vehicle by tracking the discriminant features and finds that the vehicle is in operation. The personnel monitoring unit in step S4 determines that there are no personnel in the dangerous area and there is no dangerous behavior at present.

[0072] Figure 7 This image shows the detection software used to detect a vehicle entering a mine station. A person is illegally getting off the vehicle. The vehicle tracking unit analyzes the vehicle and detects a pedestrian in the danger zone. The system triggers an alarm, marks the offending person's location, and saves the image.

[0073] The above experiments can prove the feasibility and effectiveness of the method disclosed in the present invention.

Claims

1. A method for detecting safe operation of vehicles in coal mine transport tunnels, characterized in that: The following steps are involved: Step S1: Collect data from the monitoring site to train the target detection model, delineate a polygonal electronic fence according to the specific situation, and set scene detection parameters. The specific steps include the following: Step S1.1: Select a camera position that can clearly capture the boarding and alighting locations and deploy image acquisition equipment; Step S1.2: Simulate the scene of people and vehicles running in the tunnel and collect sample images; Step S1.3: Label the car and pedestrian objects in the image and export the data required for training the object detection model; Step S1.4: Using the object detection network as the basic framework, transfer learning is performed on the training data to obtain the boarding and alighting detection model; Step S1.5: Set the car confidence threshold based on the model detection effect and pedestrian confidence threshold ,Detection results below the set confidence threshold will be discarded; Step S1.6: Based on the collected images, the track and vehicle travel range are delineated to set a polygonal danger zone, and the absolute position sequence of the vertices forming the polygon in the image is recorded; Step S2: Obtain images captured by a camera placed near a parking spot for vehicles in a lane, and use a target detection model to identify and process the images to obtain location information of the vehicle compartment and pedestrians; Step S3: Analyze the carriage position information through the carriage tracking unit to obtain the running status of the level tunnel man-car; Step S4: Analyze the position of pedestrians through the personnel monitoring unit to obtain information about people in the dangerous area, which specifically includes the following steps: Step S4.1: Filter out pedestrians whose confidence level is not less than the pedestrian confidence threshold Pedestrian detection results; Step S4.2: Determine anchor points based on pedestrian objects Intercept the pedestrian detection frame to obtain the pedestrian object judgment frame, and calculate the area of the pedestrian object judgment frame as follows: ,in is the horizontal interception starting percentage, is the horizontal interception end percentage, is the starting percentage of vertical interception, To capture the end percentage of the longitudinal direction, the pedestrian object determination anchor point is manually adjusted according to the on-site conditions of boarding and alighting; Step S4.3: Use the polygonal electronic fence to intersect with the pedestrian object judgment box, calculate the area of the intersection part and obtain the area of the pedestrian object judgment box intruding into the danger zone. ; Calculate the ratio of the area of the pedestrian object judgment box invading the dangerous area to the area of the pedestrian object judgment box itself ; Step S4.4: If the ratio R of the area of the pedestrian object judgment frame invading the dangerous area to the area of the pedestrian object judgment frame itself is greater than or equal to the intrusion ratio threshold The pedestrian is judged to be in the dangerous area, where the intrusion ratio threshold Manual adjustment according to on-site conditions; Step S5: When a pedestrian is detected in a dangerous area while the tunnel vehicle is in operation, it is determined that there is a safety hazard in the tunnel vehicle, an alarm is triggered, and an image is captured at the time of the alarm, the position of the pedestrian is marked and saved for inspection.

2. The method for detecting safe operation of vehicles in coal mine transport tunnels according to claim 1, characterized in that: The carriage tracking unit specifically includes the following steps: Step S3.1: Maintain a carriage position record table, which stores the position information of each tracked carriage in the latest N frames of images , where N is the number of retained frames, The position information quadruple of the carriage that appears for the nth time in the carriage tracked in section i; the position information quadruple consists of four values Composition, of which are the horizontal and vertical coordinates of the upper left corner of the carriage detection frame, are the horizontal and vertical coordinates of the lower right corner of the car detection frame respectively; Step S3.2: Filter out the compartments whose confidence level is not less than the confidence level threshold The carriage inspection results; Step S3.3: Take out one of the carriage detection results and calculate the distance between it and the last position of each carriage in the record table based on the position information , where j is the number of carriages in the current carriage position record table, the carriage distance is the Euclidean distance, and the distance between carriages is the distance between carriages. Location information quadruple , and the position information quadruple of carriage b , the method to calculate the distance d between them is: , Find the minimum distance ,like Not less than the set tolerance range , then put the carriage detection result into the carriage tracking record corresponding to the minimum distance, otherwise add the carriage detection result as a new carriage tracking record to the carriage position record table; Step S3.4: Repeat step S3.3 until all the filtered carriage detection results are updated; Step S3.5: Check each carriage tracked in the carriage position record table. If there is any carriage whose total movement distance in the latest N retained frames exceeds the minimum movement distance, It is determined that the level tunnel vehicle is running, where N is the number of retained frames set manually.

Images