Metal mine stope personnel safety behavior recognition method and system

Abstract

A metal mine stope personnel safety behavior recognition method and system, comprising: acquiring metal mine stope video streams of each camera; inputting the video stream of each camera into a safety behavior recognition model to obtain detection data triples; configuring a preset number of sliding windows for each camera, and caching the monitoring data triples output by the safety behavior recognition model in the sliding windows in timestamp order; judging whether the number of monitoring data triples with "illegal" wearing state data accumulated in the sliding window reaches a preset threshold; if yes, triggering an alarm, and storing the monitoring data triples with "illegal" wearing state data in the sliding window as illegal evidence; the above technical scheme realizes full-automatic detection of wearing compliance of protective equipment of personnel in a metal mine stope, illegal behavior identification and automatic storage of illegal evidence, strengthens safety supervision efficiency of the stope, reduces operation risk, and protects personnel life safety and continuous and stable operation of mine production.

Description

Technical Field

[0002] This application relates to the field of safety monitoring technology in metal mining sites, specifically to a method and system for identifying the safe behavior of personnel in metal mining sites. Background Technology

[0004] In the field of safety monitoring in metal mining sites, traditional methods of monitoring personnel safety behavior mainly rely on manual observation. Safety managers check surveillance camera footage in real time to determine whether workers are wearing safety helmets, masks, and protective clothing as required. This method is entirely dependent on manual operation, which is not only inefficient but also susceptible to subjective factors such as worker fatigue and lack of concentration, leading to significant deviations in monitoring results and making it difficult to effectively control the safety behavior of mining personnel.

[0005] While existing technologies attempt to use traditional image processing algorithms to detect whether workers in mining areas are wearing protective equipment as required—for example, using background subtraction and contour detection algorithms to process surveillance video images, extracting the contour information of workers, and analyzing changes in contour shape and position to identify movement trajectories and body postures, thereby determining whether protective equipment is being worn—these traditional algorithms have stringent image quality requirements and are highly susceptible to interference from objective factors such as sudden changes in lighting and complex background environments in the mining area. They suffer from drawbacks such as low monitoring accuracy, poor environmental adaptability, and weak anti-interference capabilities, making their detection performance insufficient to meet the actual safety monitoring needs of metal mining operations. Summary of the Invention

[0007] In view of the above problems, this application provides a method for identifying the safety behavior of personnel in metal mines, which solves the problems of low accuracy, poor environmental adaptability and weak anti-interference ability of existing monitoring.

[0008] To achieve the above objectives, the inventors provide a method for identifying the safety behaviors of personnel in metal mines, which includes the following steps:

[0009] Acquire video streams from each camera at the metal mine;

[0010] The video stream from each camera is input into the security behavior recognition model to obtain a detection data triplet; the monitoring data triplet includes the camera identifier, timestamp, and detection target data.

[0011] Each camera is configured with a sliding window with a preset number of frames, and the monitoring data triples output by the security behavior recognition model are cached in the sliding window according to the timestamp order;

[0012] Determine whether the number of monitoring data triplets with "violation" cumulative wearing status data within the sliding window reaches a preset threshold; if so, trigger an alarm and store the monitoring data triplets with "violation" wearing status data within the sliding window as evidence of violation.

[0013] Furthermore, a long-lived connection to the Redis database is established using connection pooling technology. The RTSP address corresponding to each camera is extracted from the Redis database. Then, an independent processing thread is allocated to each camera by establishing a thread pool. Based on the extracted RTSP address, the video stream of the metal mine is obtained from the camera.

[0014] Furthermore, in the step of acquiring the video stream of the metal mine from the camera in real time based on the extracted RTSP address, a connection detection and reconnection mechanism is executed simultaneously. If a frame reading failure occurs, the corresponding resources are automatically released.

[0015] Furthermore, the construction of the security behavior recognition model includes the following steps:

[0016] The training dataset consists of images containing the wearing status of miners' protective equipment under different lighting conditions, scene environments, and working postures.

[0017] The images in the training dataset are labeled with the wearing status, and the labeling categories are divided into violations and compliance;

[0018] A target detection model is established. The training dataset is input into the target detection model for training. The difference between the output and the labels of the training data is calculated using the loss function. A learning rate decay strategy is set. Through multiple rounds of forward and backward propagation, the parameters of the model are continuously optimized until the detection accuracy and recall rate reach the preset threshold. The training is then completed and saved to obtain the safety behavior recognition model.

[0019] Furthermore, the monitoring data triples output by the security behavior recognition model are cached in real time in a Redis message queue according to the timestamp order, and LPUSH or RPOP is used to ensure the timeliness of the data.

[0020] Furthermore, evidence of violations is stored on a local server or device, and the absolute address of the stored information is stored in a Redis database.

[0021] Furthermore, a storage period is set for evidence of violations stored on a local server or device, and expired data is deleted periodically at preset times.

[0022] A safety behavior identification system for personnel in a metal mine includes:

[0023] Video input layer: Connects to each camera to obtain the video stream of the metal mine.

[0024] Target detection layer: includes a safety behavior recognition model, which is used to input the video streams of the metal mine from each camera into the safety behavior recognition model for detection;

[0025] Behavioral decision layer: Determine whether the number of monitoring data triplets with "violation" cumulative wearing status data in the sliding window has reached a preset threshold; if so, trigger an alarm and store the monitoring data triplets with "violation" wearing status data in the sliding window as evidence of violation in the data storage layer;

[0026] Data storage layer: Stores evidence of violations.

[0027] Furthermore, the video input layer accesses the camera via the RTSP protocol.

[0028] Furthermore, it also includes an update and maintenance layer that automatically cleans up evidence of violations on a regular basis.

[0029] Unlike existing technologies, the above-mentioned technical solution uses a safety behavior recognition model to detect the video streams of each camera in real time. It can accurately and efficiently determine the wearing status of protective equipment by workers and output a monitoring data triplet containing camera identifiers, timestamps, and detection target data. This enables precise location of the corresponding equipment, time, and behavior information for violations. At the same time, it adopts a sliding window cumulative judgment mechanism to effectively filter out instantaneous environmental interference such as dust obstruction and sudden changes in lighting, as well as false alarms caused by single-frame detection misdetections. This significantly improves the accuracy and anti-interference robustness of violation identification. Once the alarm threshold is met, an alarm is triggered, and the corresponding violation data in the sliding window is stored as evidence. This provides a time-series coherent, complete, and verifiable basis for subsequent violation tracing, verification, and responsibility determination. Thus, it achieves fully automated detection of the compliance of personnel wearing protective equipment in metal mines, intelligent identification of violations, and automatic storage of violation evidence. This strengthens the effectiveness of mine safety supervision, reduces operational safety risks, and ensures the safety of personnel and the continuous and stable operation of mine production.

[0030] The above description of the invention is merely an overview of the technical solution of this application. In order to enable those skilled in the art to better understand the technical solution of this application and to implement it based on the description and drawings, and to make the above-mentioned objectives and other objectives, features and advantages of this application easier to understand, the following description is provided in conjunction with the specific embodiments and drawings of this application. Attached Figure Description

[0032] The accompanying drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of specific embodiments of the present invention and other related contents, and should not be considered as limitations on this application.

[0033] In the accompanying drawings of the instruction manual:

[0034] Figure 1 This is a flowchart illustrating the method for identifying the safe behavior of personnel in a metal mine, as described in a specific implementation.

[0035] Figure 2 This is a schematic diagram of the metal mine personnel safety behavior recognition system module as described in a specific implementation. Detailed Implementation

[0037] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this application in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this application and are therefore intended to limit the scope of protection of this application.

[0038] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0039] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0040] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0041] In this application, terms such as “first” and “second” are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual quantity, hierarchy or order relationship between these entities or operations.

[0042] Without further limitations, the use of terms such as “comprising,” “including,” “having,” or other similar open-ended expressions in this application is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0043] As understood in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceeding" are understood to exclude the stated number; expressions such as "above," "below," and "within" are understood to include the stated number. Furthermore, in the description of the embodiments in this application, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times," unless otherwise explicitly specified.

[0044] In the description of the embodiments of this application, the space-related expressions used, such as "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "vertical," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiments or drawings. They are only for the purpose of describing the specific embodiments of this application or for the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0045] The processor described in the embodiments of this application can be implemented by hardware, firmware, software, or a combination thereof. It can be a circuit, one or more of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field-programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, or a microprocessor. It also includes other physical, biological, or chemical structures that can implement the same or equivalent functions as the processors listed above, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some or all of the steps in the computer program or method involved in the various embodiments of this application, or any combination of the steps mentioned therein.

[0046] The computer program involved in the embodiments can be stored in a computer device readable storage medium, including but not limited to disks, magnetic tapes, magnetic cards, floppy disks, flash memory, optical disks, optical cards, read-only memory (ROM), random access memory (RAM), erasable programmable ROM (EPROM), and electrically erasable programmable ROM (EEPROM), etc., as well as other biological, physical, or chemical structures that can achieve the same or equivalent functions as the storage media listed above, such as DNA, RNA, proteins, and other units with information storage capabilities. In specific embodiments, the storage medium involved can be one of the above-mentioned media types or a combination of the above-mentioned media types. In different embodiments, the computer program involved in the embodiments can be centrally stored in a single medium or distributed across multiple media. The memory containing the computer device readable storage medium can be non-volatile memory or random access memory. These computer device readable storage media can be built into the device or connected to the device involved in the embodiments as an external device or part of an external device. In some embodiments, the memory having a computer device readable storage medium is deployed locally; in other embodiments, the memory may be deployed remotely from the processor, for example, as a network-attached memory accessed via RF circuitry or an external port and a communication network, wherein the communication network may be the Internet, one or more intranets, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), or a suitable combination thereof, as long as computer device access to the memory is enabled. Furthermore, the computer program involved in the embodiments may be stored in plaintext / ciphertext form, or it may be designed as training data, integrated and recombined through model training and implicitly stored in the parameter states of a deep neural network or other machine learning model.

[0047] A method for identifying the safe behavior of personnel in metal mines utilizes a safety behavior recognition model to perform real-time detection of video streams from various cameras. This model can accurately and efficiently determine the wearing status of protective equipment (PE) by workers, outputting a monitoring data triplet containing camera identifiers, timestamps, and detection target data. This enables precise location of the corresponding equipment, time, and behavior information for violations. Simultaneously, a sliding window cumulative judgment mechanism effectively filters out instantaneous environmental interference such as dust obstruction and sudden changes in lighting, as well as false alarms caused by single-frame detection errors, significantly improving the accuracy and robustness of violation identification. Once an alarm threshold is met, an alarm is triggered, and the corresponding violation data within the sliding window is stored as evidence. This provides a sequential, complete, and verifiable basis for subsequent violation tracing, verification, and liability determination. Therefore, this method achieves fully automated detection of the compliance of PPE wearing by personnel in metal mines, intelligent identification of violations, and automatic storage of violation evidence. This strengthens the effectiveness of mine safety supervision, reduces operational safety risks, and ensures personnel safety and the continuous and stable operation of mine production.

[0048] The following combination Figure 1 An implementation method for identifying the safety behaviors of personnel in a metal mine is provided, comprising the following steps:

[0049] Acquire video streams from each camera at the metal mine;

[0050] The video stream from each camera is input into the security behavior recognition model to obtain a detection data triplet. The monitoring data triplet includes a camera identifier, a timestamp, and detection target data. The detection target data includes wearing status data, which indicates whether the wearing status data is in violation or acceptable.

[0051] Each camera is configured with a sliding window with a preset number of frames, and the monitoring data triples output by the security behavior recognition model are cached in the sliding window according to the timestamp order;

[0052] Determine whether the number of monitoring data triplets with "violation" cumulative wearing status data within the sliding window reaches a preset threshold; if so, trigger an alarm and store the monitoring data triplets with "violation" wearing status data within the sliding window as evidence of violation.

[0053] The video streams from the aforementioned cameras in the metal mine can be acquired in various ways depending on the mine network environment, operational scenario, equipment type, and usage requirements. These methods include, but are not limited to, RTSP real-time streaming, edge gateway forwarding, local video files, and camera-specific SDK interfaces. In some embodiments, RTSP real-time streaming is used. The video streams are obtained from the RTSP addresses of each camera via the mining network based on the RTSP real-time streaming protocol. Specifically, a long-lived connection to a Redis database can be established using a connection pool. The RTSP addresses corresponding to each camera can be extracted from the Redis database. Then, a thread pool can be established to allocate independent processing threads for each camera, and the video streams from the camera in the metal mine can be obtained based on the extracted RTSP addresses.

[0054] Metal mining sites present unique environments characterized by high dust levels, easily obstructed signals, and complex equipment operating conditions. These environments are highly susceptible to video stream interruptions that are difficult to recover automatically, and can also lead to frame read failures. If allocated system resources such as threads and memory are not released promptly after a frame read failure, continuous resource occupation can occur, potentially causing thread pool exhaustion, memory overflow, and other serious problems, impacting overall system stability. Therefore, when acquiring real-time video streams from metal mining sites using extracted RTSP addresses, a connection detection and reconnection mechanism can be implemented simultaneously. If a frame read failure occurs, the corresponding resources are automatically released. This approach reduces video stream interruptions caused by environmental interference and momentary equipment failures through real-time detection and automatic reconnection, ensuring continuous safety monitoring. It also prevents faulty equipment from continuously occupying resources, thus avoiding thread pool exhaustion and memory overflow, ensuring overall system stability. Each camera has an independent thread for frame-level verification. After each frame is received, a link verification is triggered and a global connectivity test is performed once per second. When a network interruption, frame timeout, protocol response error, or thread blocking is detected, the abnormal connection is immediately terminated and a reconnection is initiated. When a frame reading fails, the entire process of thread return, connection closure, memory clearing, and log recording is completed to release resources.

[0055] The construction of the security behavior recognition model includes the following steps:

[0056] The training dataset consists of images containing the wearing status of miners' protective equipment under different lighting conditions, scene environments, and working postures.

[0057] The images in the training dataset are labeled with the wearing status, and the labeling categories are divided into violations and compliance;

[0058] A target detection model is established. The training dataset is input into the target detection model for training. The difference between the output and the labels of the training data is calculated using the loss function. A learning rate decay strategy is set. Through multiple rounds of forward and backward propagation, the parameters of the model are continuously optimized until the detection accuracy and recall rate reach the preset threshold. The training is then completed and saved to obtain the safety behavior recognition model.

[0059] The aforementioned lighting conditions include normal lighting, low light, infrared light at night, and direct strong light. The aforementioned environmental scenarios include light dust, heavy dust obstruction, water mist, and no interference. The aforementioned work postures include standing, bending over, walking, climbing, and carrying. The aforementioned protective equipment includes safety helmets, with a focus on whether they are worn correctly; of course, depending on the needs, it can include any one or more of the following: masks, safety clothing, mining lamps, and self-rescue devices. Violations of the aforementioned wearing status markings include not wearing protective equipment and wearing protective equipment improperly.

[0060] To enhance the diversity of the dataset and the generalization ability of the model, some embodiments perform data augmentation processing on the acquired images, such as random cropping, brightness adjustment, and noise addition. And / or preprocessing is performed on the labeled training dataset, including uniform image scaling (to adapt to model input requirements), random cropping, flipping, brightness and contrast adjustment, blur removal, and invalid sample removal.

[0061] The aforementioned target detection model can be a YOLO series model (such as YOLOv8, YOLO11, etc.) or a Faster R-CNN model that combines detection accuracy and real-time performance. Preferably, the target detection model is YOLO11. YOLO11 employs multi-scale feature fusion technology, constructing a multi-scale feature pyramid to extract and fuse features at different levels from the input video frame image. This allows for accurate capture of shallow detail features of targets such as miners' bodies and protective equipment, while also effectively aggregating deep semantic features. This addresses the challenge of feature extraction in harsh environments such as low light (e.g., dim lighting underground, nighttime operations) and high dust interference (e.g., dust obscuring targets, lens blurring due to dust accumulation) in metal mining sites, significantly improving the model's adaptability and detection accuracy in complex scenarios. Furthermore, this technology enhances the model's robustness to changes in target scale and partial occlusion, ensuring stable and accurate identification of protective equipment wearing status even in various working postures such as standing, bending, and climbing, meeting the real-time, efficient, and reliable safety behavior recognition needs of metal mining sites.

[0062] The monitoring data triplet, which includes device identifier, timestamp, and target data, enables the mapping of detection results to physical device space and time, facilitating the rapid location of violations. It also generates detection records with confidence weights, making it easy to judge the effectiveness of model updates based on logs.

[0063] The aforementioned monitoring data triplet includes a camera identifier, a timestamp, and detection target data. The camera identifier is a unique code for the monitoring camera, ensuring that each monitoring data triplet can be accurately traced to the specific monitoring location in the mining area. The timestamp is the actual acquisition time of each frame in the video stream. The detection target data includes the current frame image and the wearing status, marking the violation type while detecting violations. It may also include the number of people in the current frame, credibility weight, etc., ensuring that the monitoring data triplet for each frame accurately reflects the on-site safety status of that frame.

[0064] Each camera is configured with a sliding window containing a preset number of frames. This preset number of frames can be flexibly adjusted according to the pace of the data collection operation and the real-time requirements of the detection, for example, set to 5, 8, or 10 frames. This sliding window serves as a dedicated data cache unit for the corresponding camera, used only to store the monitoring data triplets output by that camera. After the safety behavior recognition model completes the detection of a single frame of video stream, it generates a monitoring data triplet. The system writes this monitoring data triplet into the corresponding camera's sliding window in real time, according to the ascending timestamp order (i.e., the acquisition of video frames). Simultaneously, the sliding window follows a "first-in, first-out" dynamic update rule. When a new monitoring data triplet is written, if the number of data frames in the window reaches the preset number, the earliest stored monitoring data triplet will be automatically removed, ensuring that the window always caches the latest consecutive preset number of monitoring data triplets. This provides a temporally coherent, complete, and effective data foundation for subsequent violation accumulation judgments based on consecutive frames. Specifically, the monitoring data triples output by the security behavior recognition model can be cached in real time in a Redis message queue according to the timestamp order, and LPUSH or RPOP can be used to ensure the timeliness of the data.

[0065] After real-time caching of the monitoring data triplets within the sliding window, violation determination logic is executed for the sliding window data corresponding to each camera: For all monitoring data triplets cached in the window in timestamp order, the triplets marked "violation" in the "wearing status" field are extracted and counted. Then, the cumulative number of violations is compared with a preset threshold to determine whether the cumulative number of violations is greater than or equal to the preset threshold. If the determination result is yes, it indicates that the miner within the monitoring range of the camera has a continuous violation of protective equipment wearing regulations, rather than a false violation caused by a single frame misdetection or momentary interference. An alarm mechanism is immediately triggered, and all monitoring data triplets with the wearing status "violation" within the sliding window are automatically extracted and stored as evidence of violation. The preset threshold can be dynamically configured according to the safety management requirements of the mining area, the risk level of the work scenario, and the model detection accuracy. For example, it is preset to 5 frames for regular work areas and 3 frames for high-risk work areas to balance alarm sensitivity and false alarm control. The alarm can include one or more of the following: local audible and visual alarm, pop-up alarm on the backend management platform, and SMS / APP push alarm for relevant personnel. In some embodiments, when pushing an alarm, a violation certificate is sent simultaneously.

[0066] Evidence of violations can be stored using random or standardized naming rules (e.g., "camera identifier_timestamp_violation frame number"). This evidence can be stored on a local server or a designated storage path on the device, forming a complete and verifiable chain of evidence that provides reliable data support for subsequent security monitoring and enforcement, as well as for handling violations. In some embodiments, the absolute storage address is also stored in a Redis database. When evidence needs to be retrieved, the local storage address of the evidence can be quickly found through Redis, allowing for direct extraction without traversing the local storage directory, thus improving evidence retrieval efficiency.

[0067] In some embodiments, in order to optimize the local storage resource usage, ensure the long-term stable operation of the system, and take into account the rationality of tracing evidence of violations and data management, a clear storage period is set for evidence of violations stored on the local server or device. The storage period can be flexibly configured according to the mining site safety management specifications, data tracing requirements and storage device capacity (for example, it is preset to 90 days in normal operation scenarios, and can be extended to 180 days in high-risk operation areas or key supervision areas). The system will establish an automatic lifecycle management mechanism based on the timestamps generated by the data. The background scheduled task will perform expired data cleanup operations according to a preset time period (preferably the low-load period from 00:00 to 02:00 every day): During execution, it will traverse the evidence of violations in the local storage directory, extract the timestamp information corresponding to each file, calculate the difference with the current system time, and if the difference exceeds the preset storage period, it will automatically perform file deletion operations, and at the same time update the data storage log to record the camera identifier, time range and number of files of the deleted files. For evidence data of major violations that need to be retained for a long time, the "permanent retention" attribute can be manually marked in the background to make it unaffected by the automatic cleanup mechanism. This ensures that while meeting the needs of daily data management, it avoids expired data occupying too many storage resources and causing system performance degradation, and also ensures the long-term traceability of key evidence of violations.

[0068] This application also provides a safety behavior recognition system for personnel in metal mines, which achieves intelligent detection and supervision of the entire process of compliance with protective equipment wearing by mine personnel through a layered architecture design. The following is combined with... Figure 2 This paper provides an implementation method for a safety behavior recognition system for personnel in a metal mine. Specifically, it includes:

[0069] Video input layer: Connects to each camera to obtain the video stream of the metal mine.

[0070] Target detection layer: includes a safety behavior recognition model, which is used to input the video streams of the metal mine from each camera into the safety behavior recognition model for detection;

[0071] Behavioral decision layer: Determine whether the number of monitoring data triplets with "violation" cumulative wearing status data in the sliding window has reached a preset threshold; if so, trigger an alarm and store the monitoring data triplets with "violation" wearing status data in the sliding window as evidence of violation in the data storage layer;

[0072] Data storage layer: Stores evidence of violations.

[0073] The aforementioned video input layer serves as the system's data acquisition entry point, and can stably acquire real-time video streams from various cameras in the metal mine through diverse access methods such as RTSP, edge gateway forwarding, and wireless mesh self-organizing networks. In some embodiments, a connection detection and reconnection mechanism is executed synchronously, automatically releasing the corresponding resources when frame reading fails, ensuring the continuity of video stream transmission and the reasonable use of system resources.

[0074] The aforementioned target detection layer incorporates a built-in security behavior recognition model. It inputs the video stream acquired from the video input layer frame by frame into the model for detection, outputting camera identifiers, precise timestamps, and detected target data. This security behavior recognition model can be trained using models that combine detection accuracy and real-time performance, such as the YOLO series (e.g., YOLOv8, YOLO11) and Faster R-CNN. The YOLO11 model is preferred, as its multi-scale feature fusion technology effectively improves detection accuracy and model robustness in low-light and dusty environments.

[0075] The aforementioned behavioral decision layer configures a sliding window with a preset number of frames for each camera, caching the monitoring data triplets output by the corresponding camera in timestamp order. The preset number of frames can be flexibly adjusted according to the mining operation rhythm and real-time detection requirements, for example, set to 5, 8, or 10 frames. The sliding window follows a "first-in, first-out" dynamic update rule. When a new monitoring data triplet is written, if the number of data frames in the window reaches the preset number of frames, the earliest stored monitoring data triplet will be automatically removed, ensuring that the window always caches the latest consecutive preset number of monitoring data triplets. This provides a temporally coherent, complete, and effective data foundation for subsequent violation accumulation judgment based on consecutive frames. By counting the number of triplets marked as "violation" in the window and comparing it with a preset threshold (which can be dynamically adjusted according to the risk level of the operation scenario), if the cumulative number of violations reaches or exceeds the preset threshold, it is determined to be a continuous violation, immediately triggering an alarm. Simultaneously, all monitoring data triplets in the "violation" state in the window are extracted as complete violation evidence and pushed to the data storage layer. The alarm may include one or more of the following: local audible and visual alarms, pop-up alarms on the backend management platform, and SMS / APP push alarms for relevant personnel. In some embodiments, when pushing an alarm, a violation notice is sent simultaneously.

[0076] The aforementioned data storage layer is responsible for the secure storage and management of evidence of violations. It saves the evidence of violations to a designated local directory, while storing the absolute address of the image, the corresponding monitoring data triplet, and alarm-related information in a Redis database. This ensures the secure retention of evidence data and supports fast background querying and retrieval. It also sets differentiated storage periods for stored data and automatically cleans up expired data through scheduled tasks, optimizing storage resource usage and providing full-process technical support for the safety supervision of mining personnel.

[0077] In some embodiments, the metal mine personnel safety behavior identification system further includes an update and maintenance layer. This layer, as a core support layer ensuring optimized system storage resources and efficient data management, is primarily responsible for the periodic automatic cleanup of violation evidence stored in the data storage layer, while also considering the rationality of data traceability and the stability of system operation. Based on mine safety management regulations, violation evidence traceability requirements, and local storage capacity, the update and maintenance layer can configure differentiated storage periods for different types of violation evidence (for example, the storage period for evidence corresponding to routine violations is preset to 90 days, while it can be extended to 180 days for high-risk work areas or key monitored locations). It also incorporates a built-in timed task scheduling mechanism to automatically trigger the cleanup process according to a preset time cycle (preferably during the low-load period in the early morning each day to avoid the cleanup operation consuming system detection resources). During the cleanup operation, the update and maintenance layer iterates through the local storage directory and Redis database index of the data storage layer to extract the generation timestamp corresponding to each piece of evidence of violation. The difference between the timestamp and the current system time is calculated. If the difference exceeds the corresponding configured storage period and the evidence is not marked with the "long-term retention" attribute, a layered cleanup is automatically performed: first, the violation evidence images and corresponding monitoring data triplet files stored on the local server are deleted; then, the associated image absolute addresses, data indexes, and other information in the Redis database are deleted simultaneously to ensure the consistency between stored data and index information. After the cleanup is completed, the update and maintenance layer generates a detailed cleanup log, recording key information such as the cleanup time, the camera identifiers involved, the time range of the violation evidence, and the number of files deleted. This facilitates the management personnel to trace the data cleanup situation and releases the occupied storage resources, preventing the system storage pressure from increasing and the operating efficiency from decreasing due to data accumulation, and ensuring the long-term stable and efficient operation of the entire recognition system.

[0078] Finally, it should be noted that although the above embodiments have been described in the text and drawings of this application, this should not limit the scope of patent protection of this application. Any technical solutions that are based on the essential concept of this application and utilize the content described in the text and drawings of this application, resulting in equivalent structural or procedural substitutions or modifications, as well as the direct or indirect application of the technical solutions of the above embodiments to other related technical fields, are all included within the scope of patent protection of this application.

Claims

1. A method for identifying safe behavior of personnel in a metal mine stope, characterized by, Includes the following steps: Acquire video streams from each camera at the metal mine; The video stream from each camera is input into the security behavior recognition model to obtain a detection data triplet; the monitoring data triplet includes the camera identifier, timestamp, and detection target data. Each camera is configured with a sliding window with a preset number of frames, and the monitoring data triples output by the security behavior recognition model are cached in the sliding window according to the timestamp order; Determine whether the number of monitoring data triplets with "violation" cumulative wearing status data within the sliding window reaches a preset threshold; if so, trigger an alarm and store the monitoring data triplets with "violation" wearing status data within the sliding window as evidence of violation.

2. The method for identifying the safety behavior of personnel in metal mines according to claim 1, characterized in that, A long-lived connection to the Redis database is established using connection pooling technology. The RTSP address corresponding to each camera is extracted from the Redis database. Then, an independent processing thread is allocated to each camera by establishing a thread pool. Based on the extracted RTSP address, the video stream of the metal mine is obtained from the camera.

3. The method for identifying the safety behavior of personnel in metal mines according to claim 2, characterized in that, In the step of acquiring the video stream of the metal mine from the camera in real time based on the extracted RTSP address, a connection detection and reconnection mechanism is executed simultaneously. If a frame reading failure occurs, the corresponding resources are automatically released.

4. The method for identifying the safety behavior of personnel in metal mines according to claim 1, characterized in that, The construction of the security behavior recognition model includes the following steps: The training dataset consists of images containing the wearing status of miners' protective equipment under different lighting conditions, scene environments, and working postures. The images in the training dataset are labeled with the wearing status, and the labeling categories are divided into violations and compliance; A target detection model is established. The training dataset is input into the target detection model for training. The difference between the output and the labels of the training data is calculated using the loss function. A learning rate decay strategy is set. Through multiple rounds of forward and backward propagation, the parameters of the model are continuously optimized until the detection accuracy and recall rate reach the preset threshold. The training is then completed and saved to obtain the safety behavior recognition model.

5. The method for identifying the safety behavior of personnel in metal mines according to claim 1, characterized in that, The monitoring data triples output by the security behavior recognition model are stored in a Redis message queue in real time according to the timestamp order for caching. LPUSH or RPOP is used to ensure the timeliness of the data.

6. The method for identifying the safety behavior of personnel in metal mines according to claim 1, characterized in that, Store evidence of violations on a local server or device, and store the absolute address of the stored information in a Redis database.

7. The method for identifying the safety behavior of personnel in metal mines according to claim 6, characterized in that, Set a storage period for evidence of violations stored on a local server or device, and periodically delete expired data at preset times.

8. A system for identifying the safe behavior of personnel in a metal mine, characterized in that, include: Video input layer: Connects to each camera to obtain the video stream of the metal mine. Target detection layer: includes a safety behavior recognition model, which is used to input the video streams of the metal mine from each camera into the safety behavior recognition model for detection; Behavioral decision layer: Determine whether the number of monitoring data triplets with "violation" cumulative wearing status data in the sliding window has reached a preset threshold; if so, trigger an alarm and store the monitoring data triplets with "violation" wearing status data in the sliding window as evidence of violation in the data storage layer; Data storage layer: Stores evidence of violations.

9. The personnel safety behavior identification system for metal mines according to claim 8, characterized in that, The video input layer connects to the camera via the RTSP protocol.

10. The personnel safety behavior identification system for metal mines according to claim 8, characterized in that, It also includes an update and maintenance layer that automatically cleans up evidence of violations on a regular basis.

Images