Real-time monitoring and abnormity early warning shooting training safety system and method

Abstract

The invention relates to the technical field of anomaly monitoring, in particular to a real-time monitoring and anomaly early warning shooting training safety system and method, and the system comprises a boundary delimiting module, a position detection module, an action construction module, an anomaly recognition module and a signal output module. According to the invention, by extracting the three-dimensional coordinates of the multiple reference points and calculating the space angle, the point section boundary structure with the consistent direction is constructed, and the accurate physical recognition of the training ground boundary is realized. Based on continuous frame shoulder, elbow and wrist position changes, speed vectors and track sequences are generated, direction changes and action trends are obtained, high-density action feature capture is achieved, continuous direction change sections in tracks are analyzed, specific action types are accurately distinguished, and boundary crossing behaviors are recognized and predicted in real time in combination with track deviation trends and boundary information. According to the processing, the organization level and dynamic change response capability of the spatial behavior data are improved, and rapid perception and judgment of the border crossing risk are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of anomaly monitoring, and in particular to a shooting training safety system and method for real-time monitoring and anomaly warning. Background Art

[0002] The technical field of anomaly monitoring includes the real-time data collection and analysis of equipment operating status, environmental parameters, and operation behaviors. The core lies in establishing effective state recognition models and risk warning mechanisms. The technical field of anomaly monitoring involves three technical levels: sensor network deployment, data transmission protocol optimization, and pattern recognition algorithm design. It forms a complete technical chain in scenarios such as industrial equipment monitoring, traffic behavior management, and medical operation specifications, focusing on solving common technical problems such as unstructured data processing, multi-source information fusion, and low-latency response, and needs to simultaneously meet data collection accuracy, system robustness, and computing resource constraints.

[0003] Among them, traditional shooting training safety management refers to the process of supervising and ensuring safety for the action behaviors of training personnel, the usage status of firearms, and the changes in the shooting range environment in a closed or open shooting training venue through manual observation or based on fixed video monitoring devices. It relies on instructors or management personnel to judge the training safety status according to experience, and uses basic video monitoring means to obtain picture information to complete the basic recording and anomaly investigation of the training process. The main means it adopts include deploying cameras at fixed positions to cover the scene, using intercom devices for real-time communication, relying on manual inspections to confirm whether the action of training personnel is standard, whether the muzzle orientation is safe, and whether there are safety risks such as crossing the line behavior or misoperation.

[0004] When dealing with unstructured spatial data, the existing technology faces the problems of sparse point distribution and blurred boundaries, resulting in the inability to accurately depict the boundaries of the training ground. It is difficult to construct a closed-space reference system in dynamic behavior recognition, and there are often phenomena where crossing behaviors are not recognized in a timely manner. At the same time, in terms of the processing of personnel action trajectories, the existing system mainly relies on simple position point changes and fails to fully utilize dynamic parameters such as trajectory direction changes and velocity vectors, resulting in misjudgments and omissions of complex action behaviors. Especially in high-speed or rapid turning actions, the determination sensitivity is low. In addition, due to incomplete extraction of key nodes for action recognition and only relying on a single sensor information source, the posture analysis results lack stability, further limiting the recognition depth of behavior events. Insufficient accuracy in links such as boundary determination, trajectory recognition, and posture analysis will affect the reliability and response timeliness of safety control, increasing the difficulty of handling after an abnormal event occurs. In a training scenario, if the initial action of a person deviating from the safe area is not recognized, it may evolve into a high-risk event within a few seconds, and the delay response mechanism under the traditional technical architecture cannot meet the control requirements of high-risk scenarios. Summary of the Invention

[0005] In order to solve the technical problems existing in the prior art, an embodiment of the present invention provides a shooting training safety system and method for real-time monitoring and abnormal warning. The technical solution is as follows: On the one hand, a shooting training safety system for real-time monitoring and abnormal warning is provided, and the system includes: A boundary delineation module, which, through a millimeter-wave radar and an infrared positioning device, obtains point cloud data in the training ground in real time, extracts the three-dimensional coordinates of multiple reference points, calculates the spatial distance and angle, combines point segments with the same direction to form a boundary structure, outputs a set of training ground boundary points, and transfers it to the position detection module; A position detection module, which calls the set of training ground boundary points, uses the point cloud data to detect the position of personnel, obtains the coordinates of the shoulder, elbow, and wrist, calculates the distance and time difference between adjacent frames, generates a velocity vector, constructs a trajectory sequence, outputs an action trajectory feature set, and transfers it to the action construction module; An action construction module, which obtains the action trajectory feature set, analyzes the trajectory directions of the shoulder, elbow, and wrist, calculates the change in the angle between adjacent frames, extracts a trajectory segment with continuous direction change, identifies the action type and outputs a label, outputs an attitude recognition record, and transfers it to the abnormal recognition module; An abnormal recognition module, based on the attitude recognition record, extracts the shoulder coordinates, identifies the movement trajectory of personnel, analyzes the direction of speed change and the degree of continuous path deviation, combines with the boundary points, identifies and predicts out-of-bounds behavior, outputs an out-of-bounds event sequence, and transfers it to the signal output module.

[0006] As a further solution of the present invention, the set of training ground boundary points includes a boundary point coordinate set, a boundary segment direction set, and a boundary contour structure. The action trajectory feature set includes a three-dimensional trajectory line, a velocity vector sequence, and a boundary proximity marker. The attitude recognition record includes an action label, a trajectory direction sequence, and an attitude change pattern. The out-of-bounds event sequence includes a trajectory number, an out-of-bounds starting point, and offset duration information.

[0007] As a further solution of the present invention, the boundary delineation module includes: A data acquisition sub-module, which, through a millimeter-wave radar and an infrared positioning device, obtains point cloud data in the training ground in real time, performs data fusion by combining time synchronization and spatial registration, identifies a point cluster area where the change amplitude of the reflection signal intensity in each frame is less than a set floating interval within a continuous time period, filters the point cluster clusters located in the observation area based on the device position label, and generates a per-frame point cluster data volume; The set floating interval refers to setting an upper and lower limit range for the signal intensity change of the same spatial point in continuous sampling frames; The coordinate calculation sub-module selects multiple consecutive reference points in each frame based on the amount of data in the frame point clusters, constructs a three-dimensional coordinate vector group according to the device positioning number, performs coordinate system conversion according to the ICP algorithm, calculates the spatial distance value between point pairs and the included angle value of three-point combinations, screens and eliminates point groups that do not conform to the continuous change interval of the included angle, and obtains the stable direction coordinate value interval; The ICP algorithm realizes the coordinate system conversion between point clusters and the screening of the stable direction interval by setting the point pair matching distance threshold, the error convergence standard and the initial positioning number; The continuous change interval of the included angle refers to the allowable fluctuation range of the change value of the vector included angle formed by multiple reference points between consecutive frames in three-dimensional space; The boundary generation sub-module selects point segments with a direction change rate lower than the direction consistency threshold according to the stable direction coordinate value interval and performs structural combination, sorts the paths according to the direction consistency of the point segments, extracts path vertices according to the Delaunay triangulation algorithm to construct a boundary structure, and generates a training ground boundary point set; The direction consistency threshold is set by statistically analyzing the distribution characteristics of the direction change rate in the stable direction coordinate interval; The Delaunay triangulation algorithm constructs a boundary vertex set according to point segments with a low direction change rate, and generates a continuous and consistent training ground boundary structure through the circumcircle legality and boundary constraints.

[0008] As a further solution of the present invention, the position detection module includes: The attitude coordinate extraction sub-module calls the three-dimensional coordinates of the boundary points in the training ground boundary point set, uses the millimeter wave radar and the infrared positioning device to detect the presence state of the person within the boundary range, and real-time collects the human point cloud data, identifies the three-dimensional spatial coordinates of the person's shoulders, elbows, and wrists in each frame, integrates them into a continuous frame coordinate sequence according to the time sequence, and generates a three-dimensional attitude coordinate sequence; The speed trajectory calculation sub-module calculates the distance between adjacent frames and the corresponding time interval based on the continuous coordinate points of the shoulders, elbows, and wrists in the three-dimensional attitude coordinate sequence, constructs a speed vector, and arranges it into a trajectory speed sequence according to the time sequence; The action feature generation sub-module extracts the movement direction according to the speed change trend of each segment according to the trajectory speed sequence, analyzes the change of the direction vectors of the shoulders, elbows, and wrists in consecutive frames, calculates the included angle offset range, identifies the continuously changing segments of the trajectory and organizes them into a behavior feature sequence, and outputs an action trajectory feature set.

[0009] As a further solution of the present invention, the action construction module includes: The trajectory extraction sub-module obtains the position information of the shoulder, elbow, and wrist in the action trajectory feature set, organizes the trajectory data into frames according to the time series, filters out the continuous frame segments of the signals and uses them as action intervals, calculates the connection direction vectors of the three points in each frame according to the coordinate changes and performs normalization processing, establishes a trajectory vector set in each frame, and generates the sequence value of the part vector change; The continuous frame segment of the signal refers to the frame sequence in which the coordinate data of the shoulder, elbow, and wrist do not have missing interruptions in multiple adjacent frames in the time series; The direction analysis sub-module, based on the sequence value of the part vector change, calculates the vector included angle values between the three points in adjacent frames, obtains the change amplitude of the included angle between frames by using the cosine similarity, filters out the frame segments with the included angle change volatility lower than the direction stability threshold, extracts the continuously changing time segments within the frame segments, and obtains the direction stable trajectory interval; The specific formula for obtaining the change amplitude of the included angle between frames by using the cosine similarity is: ; Calculate the included angle perturbation fusion value; Among them, represents the included angle perturbation fusion value of the i-th frame, represents the dot product value of the shoulder-elbow direction vector and the elbow-wrist direction vector of the i-th frame, represents the product of the moduli of the above two vectors, represents the displacement distance of the shoulder coordinates in the three-dimensional space from the i-th frame to the i + 1-th frame, represents the difference between the included angles of the elbow-wrist vectors between the i-th frame and the i + 1-th frame, is the displacement perturbation adjustment factor, is the trend adjustment factor, represents the number of frames with the same forward continuous included angle change trend, and i represents the serial number of the currently processed frame in the time series; The included angle change volatility is obtained by calculating the standard deviation of the difference sequence of the vector included angles between adjacent frames; The direction stability threshold is obtained by collecting multiple segments of stable action trajectories, calculating the standard deviation distribution of the included angle change volatility, and selecting the high quantile value as the threshold; The action recognition sub-module, according to the direction stable trajectory interval, analyzes the angle combination pattern between the shoulder, elbow, and wrist vectors in the trajectory segment, identifies the action type of the person in the training field, outputs the action type label, detects the abnormal action type, and generates the posture recognition record sequence.

[0010] As a further solution of the present invention, the abnormal recognition module includes: The moving trajectory extraction sub-module extracts the three-dimensional coordinates of the shoulders of the person in consecutive frames based on the pose recognition record, arranges the coordinate points in chronological order, identifies the spatial position change trend between consecutive coordinates, generates the moving trajectory of the shoulders in the training area, and generates a moving trajectory sequence; The path analysis sub-module arranges the coordinate points in chronological order according to the moving trajectory sequence, constructs a trajectory direction sequence based on the coordinate change trend, calculates the angular change between adjacent direction vectors, extracts the continuously changing angular trajectory segments, extends based on the end direction vector of the trajectory segment to generate a predicted moving path sequence, and generates a set of path offset feature segments; The out-of-bounds prediction sub-module, according to the set of path offset feature segments, combines the moving trajectory with the set of boundary points, compares the spatial position relationship between the path and the boundary, filters out the path segments that overlap with the boundary, extracts the person number and time position corresponding to the overlapping segments, and generates an out-of-bounds event sequence.

[0011] As a further aspect of the present invention, the specific formula for calculating the angular change between adjacent direction vectors is: ; Calculate the adjacent direction angle value; where, represents the direction angle value between the j-th frame and the (j + 1)-th frame, represents the dot product result of the direction vectors of the j-th frame and the (j + 1)-th frame, represents the adjustment coefficient of the direction difference of the j-th frame, represents the position offset between the j-th frame and the (j + 1)-th frame, represents the product of the magnitudes of the direction vectors of the j-th frame and the (j + 1)-th frame, represents the smoothing factor of the direction of the j-th frame, represents the index of the current frame.

[0012] As a further aspect of the present invention: The signal output module obtains the corresponding number and time position in the out-of-bounds event sequence, locates the abnormal path using the boundary point cloud set, combines the recognition result of the abnormal action type, outputs the warning content, sends a warning prompt message and outputs the control instruction for the acoustic and optical warning device, and generates an abnormal behavior response record; The control instruction for the acoustic and optical warning device sets different frequencies of acoustic and optical alarms according to the length of the out-of-bounds path; The real-time abnormal behavior recording unit includes the person number, abnormal area identifier, and warning response method.

[0013] As a further aspect of the present invention, the signal output module includes: An event matching sub-module obtains the corresponding numbers and time positions in the out-of-bounds event sequence, matches the personnel identifiers corresponding to the numbers, extracts the recognition results of abnormal action types, arranges the behavior time sequence and action marker content of the numbers, and generates an event status synchronization value; A path positioning sub-module, according to the event status synchronization value, calls the spatial coordinate interval corresponding to the event time period in the boundary point cloud set, locates the spatial movement path of the personnel identifier within the corresponding time, screens the coordinate point segments where boundary crossings occur in the path and extracts the continuous spatial displacement trajectory, and obtains the out-of-bounds path position sequence; A control instruction sub-module, based on the out-of-bounds path position sequence, combines the relationship between the event time and the trajectory number, extracts the corresponding warning level type information, matches the sound and light device control parameters according to the warning level type, and generates an abnormal behavior response record.

[0014] On the other hand, a shooting training safety method for real-time monitoring and abnormal warning is provided. This method is applied to a shooting training safety system for real-time monitoring and abnormal warning. The method includes: S1: Through millimeter-wave radar and infrared positioning devices, real-time obtain the point cloud data in the training ground, extract the three-dimensional coordinates of multiple reference points, calculate the distances and direction angles between multiple reference points based on the point cloud spatial coordinates, classify the point segments with consistent direction change amplitudes, integrate them into a continuous boundary line segment path, and output it as a training ground boundary point set; S2: Obtain the training ground boundary point set, use the point cloud data to detect the three-dimensional coordinate data of personnel in the training ground, obtain the spaces of the shoulders, elbows, and wrists, calculate the distances and time differences between adjacent time frames, generate velocity vectors, and construct a continuous trajectory path of points in time series, and output it as an action trajectory feature set; S3: Obtain the action trajectory feature set, by extracting the direction angle change values between each pair of adjacent frames in the shoulder, elbow, and wrist trajectory vectors, extract the continuously changing direction trajectory segments, identify the action types and output the corresponding action labels, and output it as a posture recognition record; S4: Obtain the posture recognition record, extract the shoulder coordinate sequence, identify the personnel movement trajectory, analyze the speed change direction and the degree of continuous path deviation, and combine the boundary point positions to identify and predict out-of-bounds behaviors, and output an out-of-bounds event sequence; S5: Obtain the corresponding numbers and time positions of the out-of-bounds event sequence, use the boundary point cloud set to locate the abnormal path, combine the action label content in the corresponding posture recognition record, generate a warning content according to the time index and the abnormal path number, send a warning prompt message and output the control instruction of the sound and light device, and generate an abnormal behavior response record.

[0015] The beneficial effects brought by the technical solution provided by the embodiments of the present invention at least include: By obtaining the point cloud data of the training ground and extracting the three-dimensional coordinates of multiple reference points, combining the calculations of spatial distance and angle, combining point segments with the same direction to form a three-dimensional boundary structure, effectively improving the accuracy of the physical closure determination of complex spatial scenes and enhancing the accuracy of environmental modeling. On this basis, use the point cloud data of consecutive frames to locate key nodes such as shoulders, elbows, and wrists, calculate the velocity vector through the inter-frame distance and time difference, and then construct a dynamic feature sequence including direction changes and trajectory extensions, with the ability to express action trajectories with higher timeliness and granularity. By analyzing the angle changes between adjacent frames in the trajectory, extracting trajectory segments with continuous directions, it can efficiently identify action types and record postures, providing structured input for subsequent determination. Further, combining the recognized trajectories and action posture information, extracting the changing trend of shoulder coordinates and the continuous offset characteristics of the trajectory, and integrating the boundary structure information, identifying and predicting out-of-bounds behaviors, so as to realize the real-time prediction and timely response to out-of-bounds risks. In the overall processing logic, the coordinated cooperation of multi-point three-dimensional information extraction, multi-level trajectory analysis, and high-frequency posture change recognition significantly enhances the fine-grained capture ability of spatial action behaviors, improves the detection accuracy and response efficiency of out-of-bounds events, and realizes the accurate recognition and risk prevention and control of dynamic behaviors within the safety control boundary. Brief Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0017] Figure 1 It is the system flow chart of the present invention; Figure 2 It is the schematic diagram of the system framework of the present invention; Figure 3 It is the schematic diagram of the method steps of the present invention. Detailed Embodiments

[0018] The following will describe the technical solutions in the present invention with reference to the drawings.

[0019] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of the word "example" is intended to present concepts in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one of the two.

[0020] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same. "Of", "corresponding", and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same.

[0021] In the embodiments of the present invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meanings they express are the same.

[0022] To make the technical problems, technical solutions, and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0023] The embodiments of the present invention provide a shooting training safety system for real-time monitoring and abnormal warning. Please refer to Figures 1 to 2 , the present invention provides a technical solution. A shooting training safety system for real-time monitoring and abnormal warning includes: A boundary delineation module, which, through a millimeter-wave radar and an infrared positioning device, real-time obtains the point cloud data in the training ground, extracts the three-dimensional coordinates of multiple reference points, calculates the spatial distance and angle, combines the point segments with the same direction to form a boundary structure, outputs the training ground boundary point set and transmits it to the position detection module; A position detection module, which calls the training ground boundary point set, uses the point cloud data to detect the personnel position, obtains the coordinates of the shoulder, elbow, and wrist, calculates the distance and time difference between adjacent frames, generates a velocity vector, constructs a trajectory sequence, outputs an action trajectory feature set and transmits it to the action construction module; An action construction module, which obtains the action trajectory feature set, analyzes the trajectory directions of the shoulder, elbow, and wrist, calculates the change in the angle between adjacent frames, extracts the trajectory segments with continuous direction changes, identifies the action type and outputs a label, outputs an attitude recognition record and transmits it to the abnormal recognition module; An abnormal recognition module, based on the attitude recognition record, extracts the shoulder coordinates, identifies the personnel movement trajectory, analyzes the direction of speed change and the degree of continuous path deviation, combines with the boundary points, identifies and predicts the out-of-bounds behavior, outputs an out-of-bounds event sequence and transmits it to the signal output module; A signal output module, which obtains the corresponding numbers and time positions in the out-of-bounds event sequence, locates the abnormal path using the boundary point cloud set, combines with the recognition result of the abnormal action type, outputs the warning content, sends a warning prompt message and outputs a control instruction for the sound and light warning device, and generates an abnormal behavior response record; The control instruction for the sound and light warning device is to set sound and light alarms with different frequencies according to the length of the out-of-bounds path.

[0024] The training ground boundary point set includes a boundary point coordinate set, a boundary segment direction set, and a boundary contour structure. The action trajectory feature set includes a three-dimensional trajectory line, a speed vector sequence, and a boundary approach marker. The posture recognition record includes an action label, a trajectory direction sequence, and a posture change pattern. The out-of-bounds event sequence includes a trajectory number, an out-of-bounds start point, and offset duration information. The real-time abnormal behavior record unit includes a personnel number, an abnormal area identifier, and a warning response method.

[0025] Please refer to Figure 1 and Figure 2 , the boundary demarcation module includes: A data acquisition sub-module that, through a millimeter-wave radar and an infrared positioning device, obtains point cloud data within the training ground in real time. Combining time synchronization and spatial registration, it performs data fusion, identifies point cluster regions where the change amplitude of the reflection signal intensity within a continuous time period is less than a set floating interval, and filters out point cluster clusters located within the observation area based on the device position label to generate frame-by-frame point cluster data volumes; First, obtain the point cloud data within the training ground through a millimeter-wave radar and an infrared positioning device. The real-time sampling frequency of the device is set to collect 100 frames of data per second. The point cloud data collected in each frame includes the reflection signal intensity, a timestamp, and a position label. To ensure the high precision of the data, the device time synchronization accuracy requirement is within the microsecond level. After synchronization is completed, spatial registration is performed on the data, that is, based on the positioning information of the device, the data collected by different devices is merged into a point cloud in a unified coordinate system through an accurate spatial transformation algorithm. The specific spatial registration method is based on the linear least squares method for optimal registration, so as to ensure that the spatial position of each point in the merged point cloud has a high precision. Next, for each frame of data, according to the set floating interval, identify the point cluster region where the change amplitude of the reflection signal intensity is less than this interval. This floating interval value is obtained through historical data statistics and is set to a signal intensity fluctuation range of ±5%. Taking the signal intensity of a certain collection point as an example, assuming the current point intensity is 100 dB, the floating interval is 95 dB to 105 dB. If the change in the reflection signal intensity of a point cluster within a certain time period is less than this range, then this point cluster is considered to meet the conditions. On this basis, the device position label will be used to filter out point cluster clusters located within the observation area. For example, assuming the device position label is (x = 100, y = 200, z = 50), if the spatial position of a point cluster falls within the set observation area, then this point cluster is considered to meet the requirements. After this screening process, finally generate frame-by-frame point cluster data volumes for subsequent processing; Setting the floating interval means setting an upper and lower limit range for the signal intensity change of the same spatial point in consecutive sampled frames; The coordinate calculation sub-module selects multiple consecutive reference points in each frame based on the amount of data in the frame cluster of points, constructs a three-dimensional coordinate vector group according to the device positioning number, performs coordinate system conversion according to the ICP algorithm, calculates the spatial distance value between point pairs and the included angle value of three-point combinations, filters and eliminates point groups that do not conform to the continuously changing interval of the included angle, and obtains the stable direction coordinate value interval; In the coordinate calculation sub-module, first, select multiple consecutive reference points in each frame of data as reference coordinate points. Suppose 3 reference points are selected as A(10, 20, 30), B(15, 25, 35), and C(20, 30, 40) respectively, and the spatial positions of these points are obtained by an infrared positioning device. Then, based on the data of these reference points, construct a three-dimensional coordinate vector group according to the positioning number of the device, and use the ICP algorithm (Iterative Closest Point algorithm) for coordinate system conversion. The ICP algorithm filters out point pairs with smaller distance errors by calculating the distances between the reference points A, B, and C. The specific calculation process is as follows: first, calculate the spatial distance between point pairs. For example, calculate the distance from point A to point B, that is: , substitute the coordinates of A(10, 20, 30) and B(15, 25, 35) to get: ; If the error of this point pair meets the set threshold (for example, the set threshold is 10 units), then this point pair is considered valid. Based on these valid point pairs, calculate the spatial distance value between point pairs and the included angle value of three-point combinations. For three points A, B, and C, calculate the included angle between the two vectors AB and BC they form. For example, use the vector dot product formula to calculate the included angle: , further filter and eliminate point groups that do not conform to the continuously changing interval of the included angle. The continuously changing interval of the included angle is obtained by statistically analyzing the range of changes in the included angles between reference points in multiple consecutive frames. Suppose the set range of included angle changes is ±10°. If the included angle change between consecutive frames exceeds this range, then this point group is eliminated, and finally, the stable direction coordinate value interval is obtained; The ICP algorithm realizes coordinate system conversion between point clusters and screening of stable direction intervals by setting the point pair matching distance threshold, error convergence criterion, and initial positioning number; The continuously changing interval of the included angle refers to the allowable fluctuation range of the change value of the included angle of vectors formed by multiple reference points between consecutive frames in three-dimensional space.

[0026] The boundary generation sub-module selects point segments with a direction change rate lower than the direction consistency threshold according to the stable direction coordinate value interval and performs structural combination, sorts the paths according to the direction consistency of the point segments, extracts path vertices according to the Delaunay triangulation algorithm to construct a boundary structure, and generates a set of training ground boundary points; In the boundary generation sub-module, first, based on the direction stability coordinate value range, point segments with a direction change rate lower than the direction consistency threshold are selected for structural combination. Suppose in a certain frame of data, point segments AB and BC are selected as candidate point segments, with their direction change rates being 2° and 5° respectively, and the set direction consistency threshold being 10°. Since the direction change rates of both are lower than the threshold, these two point segments can be structurally combined. Next, the paths are sorted according to the direction consistency of the point segments to obtain the sorting results of each point segment. Taking point segments AB and BC as examples, after sorting based on direction consistency, a new point segment order is obtained, and the Delaunay triangulation algorithm is used to extract path vertices to construct the boundary structure. Suppose points A(10, 20, 30), B(15, 25, 35), and C(20, 30, 40) are sequentially selected from the corner of the training ground. According to the Delaunay algorithm, the connections of the three points form a triangle to generate the boundary structure. Finally, a set of boundary points of the training ground is obtained, and these points represent the actual boundary positions of the training ground; The direction consistency threshold is set by statistically analyzing the distribution characteristics of the direction change rate in the direction stability coordinate range; The Delaunay triangulation algorithm constructs a boundary vertex set based on point segments with a low direction change rate, and generates a continuous and consistent training ground boundary structure through circumcircle legality and boundary constraints; Please refer to Figure 1 and Figure 2 , the position detection module includes: An attitude coordinate extraction sub-module, which calls the three-dimensional coordinates of the boundary points in the set of training ground boundary points, uses the obtained millimeter-wave radar and infrared positioning devices to detect the presence status of personnel within the boundary range, and real-time collects the human point cloud data, identifies the three-dimensional space coordinates of the shoulders, elbows, and wrists of the personnel in each frame, and integrates them into a continuous frame coordinate sequence according to the time order to generate a three-dimensional attitude coordinate sequence; First, through the millimeter-wave radar and infrared positioning devices, the point cloud data within the training ground is obtained in real time. This data will include the three-dimensional coordinates of the reference points within the training ground, and based on this, the spatial layout of the boundary at each moment is judged. By analyzing the three-dimensional coordinates of the boundary points and combining the positions of the personnel within the boundary range, the coordinates of key positions such as the shoulders, elbows, and wrists of the personnel are extracted in real time using the point cloud data. This point cloud data will be calibrated at each acquisition moment, and then the three-dimensional spatial positions of the shoulders, elbows, and wrists are identified. For example, suppose at a certain moment, the coordinate of the shoulder is , the elbow is , and the wrist is , the system will integrate these data to form a set of continuous three-dimensional coordinate sequences. Based on the time sequence, these frame data will be associated one by one, and an overall three-dimensional pose coordinate sequence will be generated. Through the coordinate synthesis of consecutive frames, the movement trajectory of each key point over time can be accurately described, and finally a three-dimensional pose coordinate sequence containing multiple frame data will be output. If the coordinate at the current time point is , then the pose sequence of consecutive frames is , providing data for further motion analysis.

[0027] The speed trajectory calculation sub-module calculates the distance between adjacent frames and the corresponding time interval based on the consecutive coordinate points of the shoulder, elbow, and wrist in the three-dimensional pose coordinate sequence, constructs a speed vector, and arranges it in chronological order into a trajectory speed sequence; The core task of the module is to calculate the motion speed between adjacent frames based on the continuous shoulder, elbow, and wrist position data extracted from the pose coordinate sequence. First, obtain the three-dimensional coordinate values of the shoulder, elbow, and wrist at each moment, and then calculate the coordinate changes between adjacent frames. For example, for the displacement of the shoulder between frames and , the calculation process is , and then according to the time interval , the speed vector will be calculated by the following formula: , in the same way, the speed vectors of the elbow and wrist will also be calculated, and the corresponding speed trajectory sequences will be constructed respectively. These speed vectors will be arranged in chronological order for subsequent behavior analysis. Assume that between and , the displacement of the shoulder is meters, and the time interval is seconds, then the calculated speed vector is: m / s. Thus, the speed trajectory sequence will include the speed values of multiple frames, constituting the speed evolution information in the entire motion process for use in the next stage of motion analysis.

[0028] The motion feature generation sub-module extracts the motion direction according to the trajectory speed sequence, analyzes the change of the direction vectors of the shoulder, elbow, and wrist in consecutive frames, calculates the included angle offset range, identifies the continuously changing segments of the trajectory, and arranges them into a behavior feature sequence, outputting the action trajectory feature set; Based on the constructed trajectory speed sequence, the trend of each speed change will be further analyzed. First, extract the motion direction in each trajectory segment, analyze the change of the direction vectors of the shoulder, elbow, and wrist between frames, and calculate the included angle offset range between the direction vectors. For each segment of motion, the system will calculate the included angle change between consecutive frames. For example, assume that at two time points and Between, the shoulder direction vector is , while at it is , then the included angle will be calculated by the following formula: , in this example, the calculated included angle is . At this time, the change in the direction vector indicates a significant change in the direction of the action. Through such calculations, the module can extract the changing segments of the trajectory from the continuous coordinate and speed data, and identify specific action types based on these segments. Finally, these extracted continuously changing segments of the trajectory will be organized into a sequence of behavioral features and output as an action trajectory feature set for subsequent anomaly detection and behavior analysis.

[0029] Please refer to Figure 1 and Figure 2 , the action construction module includes: A trajectory extraction sub-module that obtains the position information of the shoulder, elbow, and wrist in the action trajectory feature set, organizes the trajectory data in frames according to the time series, filters the continuously-signaled frame segments as action intervals, calculates the connection direction vectors of the three points in each frame according to the coordinate changes and normalizes them, establishes a set of trajectory vectors in each frame, and generates the sequence values of the part vector changes; First, by obtaining the position information of the shoulder, elbow, and wrist, and organizing these coordinate data in chronological order, the trajectory data of consecutive frames is generated. The coordinates of the shoulder, elbow, and wrist at each time point will be organized into a time series in turn. After screening these data, a frame segment with continuous signals can be provided. The definition of a continuously-signaled frame segment is that in the time series, the coordinate data of the shoulder, elbow, and wrist at three points do not have any missing or interruption in multiple adjacent frames. Taking a simple example, assume that at a certain moment, the shoulder position is , the elbow position is , the wrist position is , and through calculating the coordinate changes of these data, the direction vectors and in each frame are obtained. Next, by normalizing the vectors of each frame, a normalized direction vector is obtained, and these direction vectors are combined into a set of trajectory vectors. Finally, by forming a sequence of all these trajectory vectors, the sequence of vector changes of each part is generated, and these data are used as input for subsequent action analysis.

[0030] The continuously-signaled frame segment refers to a frame sequence in which the coordinate data of the shoulder, elbow, and wrist at three points do not have any missing or interruption in multiple adjacent frames in the time series.

[0031] The direction analysis sub-module calculates the included angle values between vectors of three points in adjacent frames based on the sequence values of part vector changes, obtains the change amplitude of the included angle between frames using cosine similarity, filters out the frame segments with the included angle change volatility lower than the direction stability threshold, extracts the continuous change time segments within the frame segments, and obtains the direction stability trajectory interval; The specific formula for obtaining the change amplitude of the included angle between frames using cosine similarity is: ; Calculate the included angle perturbation fusion value; Among them, represents the included angle perturbation fusion value of the i-th frame, represents the dot product value of the shoulder-elbow direction vector and the elbow-wrist direction vector of the i-th frame, represents the product of the magnitudes of the above two vectors, represents the displacement distance of the shoulder coordinates in the three-dimensional space from the i-th frame to the i+1-th frame, represents the difference in the included angle of the elbow-wrist vector between the i-th frame and the i+1-th frame, is the displacement perturbation adjustment factor, is the trend adjustment factor, represents the number of frames with consistent forward continuous included angle change trends, and i represents the current frame number being processed in the time series; Formula: ; Detailed explanation of the formula and calculation derivation process: This formula is used to calculate the included angle perturbation fusion value of the -th frame, comprehensively considering factors such as the similarity between the shoulder-elbow direction vector and the elbow-wrist direction vector, shoulder displacement, elbow-wrist angle change, and trend adjustment. The following details each parameter in the formula and calculates with actual data; Parameter meaning and setting values: : The dot product value of the shoulder-elbow direction vector and the elbow-wrist direction vector of the -th frame, set ; : The product of the magnitudes of the shoulder-elbow direction vector and the elbow-wrist direction vector of the -th frame, ; : The displacement distance of the shoulder coordinates in the three-dimensional space from the -th frame to the -th frame, set ; : The difference in the included angle of the elbow-wrist vector between the -th frame and the -th frame, set ; : Displacement perturbation adjustment factor, usually between 0.1 and 0.5, set ; : Trend adjustment factor, usually between 0.01 and 0.1, set ; : Number of frames with consistent forward continuous angle change trend, usually between 2 and 10, set ; Substitute into the formula for calculation: Calculation process: Calculating the cosine similarity part: ; Calculating the displacement perturbation part: ; ; Calculating the trend adjustment part: ; Calculating the angle perturbation fusion value: ; This result indicates that for the frame, the angle perturbation fusion value is approximately 3.772, indicating that in this frame, the similarity between the shoulder-elbow direction vector and the elbow-wrist direction vector is relatively high, the influence of displacement perturbation and elbow-wrist angle change on angle perturbation is moderate, and the angle change trend between consecutive frames remains consistent.

[0032] The angle change volatility is obtained by calculating the standard deviation of the sequence of differences in vector angles between adjacent frames; The direction stability threshold is obtained by collecting multiple segments of stable motion trajectories, calculating the standard deviation distribution of the angle change volatility, and then selecting the high quantile value as the threshold.

[0033] The action recognition sub-module analyzes the angle combination patterns between the shoulder, elbow, and wrist vectors in the trajectory segment according to the direction stability trajectory interval, identifies the action types of the personnel in the training field, outputs the action type labels, detects abnormal action types, and generates an attitude recognition record sequence; In the action recognition sub-module, first, the selected direction stability trajectory intervals are used as the basis for analysis. These intervals have been processed by the previous sub-module to ensure continuous signals without missing values. Then, the vector combination patterns between the shoulder, elbow, and wrist in each frame are analyzed. This process includes calculating the angle between each pair of vectors to determine the posture changes during the movement. For example, assume that in a certain frame, the vector from the shoulder to the elbow is , and the vector from the elbow to the wrist is , then calculate according to the included angle between these two vectors for calculation: , through this calculation, the action features of each frame can be obtained, and through the combination of these features, the specific action types of the personnel in the training ground can be identified. If the angle change of a certain action is significantly different from the known standard postures (such as raising a gun, aiming, shooting, etc.), the system will identify it as an abnormal action. By comparing the postures and angle changes between consecutive frames, a posture recognition record sequence is generated for further abnormal action detection. According to different action types and the identified abnormal actions, this module outputs the corresponding action type labels and the abnormal action recognition results; Table 1 Trajectory Data Samples

[0034] As shown in Table 1, the coordinates of the shoulder, elbow, and wrist in the three-dimensional position data are integrated through time series, and the direction vectors are calculated. After normalization processing in each frame of data, the direction vectors are output, and the action trajectories are further generated.

[0035] Please refer to Figure 1 and Figure 2 , the abnormal recognition module includes: A moving trajectory extraction sub-module, based on the posture recognition record, extracts the three-dimensional coordinates of the shoulder of the personnel in consecutive frames, arranges the coordinate points in chronological order, identifies the spatial position change trend between consecutive coordinates, generates the moving trajectory of the shoulder in the training area, and generates a moving trajectory sequence; In the trajectory extraction sub-module, first, through the posture recognition technology, record the various posture data of the personnel in the training ground, especially the three-dimensional coordinates of the shoulder. These data are collected once per second based on the sensor system, and the spatial coordinate values at each time point are recorded. For example, the three-dimensional coordinates of the shoulder recorded at a certain moment are (x = 5.0, y = 3.2, z = 1.5), and this coordinate will be recorded synchronously with the time stamp to form a time series. Then, arrange the coordinates obtained for each frame in chronological order to ensure that the data order does not go wrong. By calculating the spatial position change trend between consecutive coordinate points, specifically, calculating the distance change between every two adjacent coordinate points, using the following formula: , assuming that the shoulder coordinates at a certain moment are (x1 = 5.0, y1 = 3.2, z1 = 1.5), and the shoulder coordinates at the next moment are (x2 = 5.2, y2 = 3.3, z2 = 1.6), then the spatial distance between the two points is: , through this method, the continuous coordinate change trend will reflect the movement trajectory of the shoulder. Combining the entire time series, a complete shoulder movement trajectory sequence is generated. For example, if the continuous coordinate sequences are (5.0, 3.2, 1.5), (5.2, 3.3, 1.6), (5.5, 3.6, 1.8), then these data form a path representing the movement trajectory of a person in the training area.

[0036] The path analysis sub-module arranges the coordinate points in chronological order according to the movement trajectory sequence, constructs a trajectory direction sequence based on the coordinate change trend, calculates the angular change between adjacent direction vectors, extracts the continuously changing angular trajectory segments, extends based on the end direction vector of the trajectory segment to generate a predicted movement path sequence, and generates a set of path offset feature segments; Formula: ; Detailed explanation of the formula and the derivation process of the formula calculation: The formula is used to calculate the included angle value of the direction vectors between the j-th frame and the j + 1-th frame, which is used to judge the direction change trend between these two frames. The result is used to extract the continuously changing angular trajectory segments to support the generation process of the predicted movement path; Parameter meanings and setting values: is the dot product value of the direction vectors of the j-th frame and the j + 1-th frame. The dot product calculation formula is: , set the current frame vector , the next frame vector , then ; is the direction difference adjustment coefficient, which takes values from 0.05 to 0.15 in stable movements. The value is set to 0.1 this time; is the position offset between the j-th frame and the j + 1-th frame, and the calculation method is the Euclidean distance of the spatial coordinate difference between the two frames, , set the acquisition coordinates as and ; then ; is the product of the moduli of the direction vectors of the two frames. The modulus of the vector is calculated as: ; is the direction smoothing factor. When the standard deviation is set to 0.03, the corresponding factor value is set to 0.03; Substitute the parameters into the formula for calculation: ; ; The result of 10.1° indicates that the change in the included angle of the direction vector between the current frame and the next frame is in a relatively continuous state. This angle change value will be compared with the set threshold for continuous angle change. If the included angle change values in consecutive frames do not exceed this threshold, the corresponding trajectory segment is extracted as the trajectory segment with continuous angle change, providing a trajectory basis for subsequent movement path prediction.

[0037] The out-of-bounds prediction sub-module, based on the set of path deviation feature segments, combines the movement trajectory and the set of boundary points, compares the spatial position relationship between the path and the boundary, filters out the path segments that overlap with the boundary, extracts the personnel numbers and time positions corresponding to the overlapping segments, and generates an out-of-bounds event sequence. In the out-of-bounds prediction sub-module, first, through the identification of path deviation feature segments, the spatial relationship between the movement trajectory of the personnel and the set of training ground boundary points is extracted. A path deviation feature segment refers to the part of the trajectory that significantly deviates from the normal expected path, usually with a large change in this part of the path. For example, if the trajectory in a certain frame moves from (x = 5.0, y = 3.2) to (x = 8.0, y = 4.0), this large deviation is considered a path deviation. Then, combined with the data in the set of boundary points, the path and the boundary are compared to determine whether there is an overlap. For example, if the boundary points are (x = 6.0, y = 4.0), (x = 8.0, y = 5.0), and a certain path segment passes through the area between these two points, then this path segment is considered to overlap with the boundary. At this time, the overlapping path segments are filtered, and the personnel numbers and time positions corresponding to the overlapping segments are recorded. For example, assume that the starting point of a certain path segment is (x = 7.8, y = 3.5), the ending point is (x = 8.2, y = 4.1), and this path segment overlaps within the set range of the training area boundary. The selected overlapping path segment will correspond to a specific personnel number (such as number 001) and a timestamp (for example, 15:32:10). In this way, an out-of-bounds event sequence is generated, recording the out-of-bounds information of the personnel, facilitating subsequent tracking and processing of out-of-bounds behaviors. Table 2 Data table of out-of-bounds prediction results

[0038] As shown in Table 2, the out-of-bounds prediction sub-module records the out-of-bounds events of personnel numbers 001 and 002. Each record contains the start time, end time, start coordinates, and end coordinates. This information can help the system quickly locate out-of-bounds behaviors and compare them with the boundary of the training area, further ensuring the safety and regularity of the training ground.

[0039] Please refer to Figure 1 and Figure 2 , the signal output module includes: The event matching sub-module obtains the corresponding numbers and time positions in the out-of-bounds event sequence, matches the personnel identifiers corresponding to the numbers, extracts the recognition results of abnormal action types, organizes the behavioral time series and action marker contents of the numbers, and generates an event status synchronization value; In the event matching sub-module, first, according to the numbers and time positions in the out-of-bounds event sequence, the detailed information of each event is obtained therefrom. For each out-of-bounds event, the system matches the corresponding personnel identifier according to the event number and extracts the recognition result of the abnormal action type of the personnel during the event period. The system associates the event number with the personnel identifier and integrates its behavioral time series and action marker contents to generate an event status synchronization value containing the detailed information of the event. For example, in a certain out-of-bounds event, assume the number is , the event time is to , the personnel identifier is , the abnormal action type is "beyond the shooting boundary", and through the action marker content and time data extracted by the system, a synchronization value is formed as the input for subsequent processing to ensure the accurate association between the personnel identifier, time position, and action type of each event.

[0040] The path positioning sub-module, according to the event status synchronization value, calls the spatial coordinate interval corresponding to the event period in the boundary point cloud set, locates the spatial movement path of the personnel identifier during the corresponding time, screens the coordinate point segments where boundary crossings occur in the path and extracts the continuous spatial displacement trajectory, and obtains the out-of-bounds path position sequence; In the path positioning sub-module, first, based on the event status synchronization value, the system locates the spatial coordinate interval corresponding to the event period by calling the data in the boundary point cloud set. By tracking the personnel movement path during this period, the system identifies the spatial path of the personnel during the event. These paths will be used to screen the coordinate point segments where boundary crossings occur. For example, assume the time period of a certain event is to , the system will screen out the coordinate data during this period from the boundary point cloud set and mark the trajectory points where boundary crossings occur. When a certain segment of the path is detected to exceed the set boundary, this segment will be extracted and marked as an out-of-bounds path. For example, in event , the coordinate segment of the personnel's path that exceeds the boundary between the time periods to is , and this part of the path is identified as an out-of-bounds path and output as the out-of-bounds path position sequence for use by the subsequent control instruction module.

[0041] The control instruction sub-module extracts the corresponding early warning level type information based on the out-of-bounds path position sequence, combines the event time and the relationship with the trajectory number, matches the sound and light device control parameters according to the early warning level type, and generates an abnormal behavior response record; The function of the control instruction sub-module is to extract the corresponding early warning level type information according to the out-of-bounds path position sequence, combine the relationship between the event time and the trajectory number, and generate the control parameters of the sound and light devices according to the early warning level. In this module, the system first extracts the early warning level type (such as "high-risk warning", "medium-risk warning", etc.) according to the event time corresponding to the out-of-bounds path position sequence, and matches the corresponding device control parameters according to the early warning level. For example, assume that in the event the out-of-bounds path position sequence is . According to the pre-set rules, when the early warning level of the out-of-bounds path is "high-risk warning", the system will match the corresponding sound and light device control parameters, such as a sound frequency of 500 Hz and a light intensity of 80%. Finally, these control parameters will generate an abnormal behavior response record, record the response operations of the devices, and provide timely early warning prompts for relevant personnel. For example, when the system identifies that the event number is a high-risk warning, the generated control instruction is "500 Hz frequency, 80% light intensity".

[0042] Please refer to Figure 3 , the method includes: S1: Through the millimeter-wave radar and the infrared positioning device, the point cloud data in the training ground is obtained in real time, the three-dimensional coordinates of multiple reference points are extracted, the distances and direction angles between multiple reference points are calculated based on the point cloud space coordinates, the point segments with the same direction change amplitude are classified, integrated into continuous boundary line segment paths, and output as the training ground boundary point set; S2: Obtain the training ground boundary point set, use the point cloud data to detect the three-dimensional coordinate data of the personnel in the training ground, obtain the spaces of the shoulders, elbows, and wrists, calculate the distances and time differences between adjacent time frames, generate velocity vectors, and construct a continuous trajectory path of points in time series, and output as the action trajectory feature set; S3: Obtain the action trajectory feature set, extract the direction angle change values between each pair of adjacent frames in the shoulder, elbow, and wrist trajectory vectors, extract the continuously changing direction trajectory segments, identify the action type and output the corresponding action label, and output as the posture recognition record; S4: Obtain the posture recognition record, extract the shoulder coordinate sequence, identify the personnel movement trajectory, analyze the speed change direction and the degree of continuous path deviation, and combine the boundary point positions to identify and predict the out-of-bounds behavior, and output the out-of-bounds event sequence; S5: Obtain the corresponding numbers and time positions of the out-of-bounds event sequences, use the boundary point cloud set to locate the abnormal path, combine the action label content in the corresponding attitude recognition record, generate warning content according to the time index and the abnormal path number, send warning prompt information and output control instructions for the sound and light devices, and generate an abnormal behavior response record.

[0043] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access, or a data storage device such as a server or data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0044] It should be understood that the term "and / or" in this article is only a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. In addition, the character " / " in this article generally represents an "or" relationship between the preceding and following associated objects, but it may also represent an "and / or" relationship, which can be specifically understood with reference to the context.

[0045] In the present invention, "at least one" means one or more, and "a plurality" means two or more. "At least one of the following" or its similar expressions refer to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or multiple.

[0046] It should be understood that in various embodiments of the present invention, the magnitudes of the serial numbers of the above processes do not imply the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0047] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Skilled professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0048] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the devices, apparatuses, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0049] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be in an electrical, mechanical, or other form.

[0050] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0051] In addition, the functional units in various embodiments of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0052] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0053] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A shooting training safety system for real-time monitoring and abnormal warning, characterized in that, The system includes: A boundary delineation module that, through a millimeter-wave radar and an infrared positioning device, real-time obtains point cloud data in the training ground, extracts the three-dimensional coordinates of multiple reference points, calculates the spatial distance and angle, combines point segments with the same direction to form a boundary structure, outputs the training ground boundary point set and transmits it to the position detection module; A position detection module that calls the training ground boundary point set, uses the point cloud data to detect the position of a person, obtains the coordinates of the shoulder, elbow, and wrist, calculates the distance and time difference between adjacent frames, generates a velocity vector, constructs a trajectory sequence, outputs the action trajectory feature set and transmits it to the action construction module; An action construction module that obtains the action trajectory feature set, analyzes the trajectory directions of the shoulder, elbow, and wrist, calculates the change in the angle between adjacent frames, extracts the trajectory segments with continuous direction changes, identifies the action type and outputs a label, outputs the posture recognition record and transmits it to the anomaly recognition module; An anomaly recognition module that, based on the posture recognition record, extracts the shoulder coordinates, identifies the movement trajectory of the person, analyzes the direction of speed change and the degree of continuous path deviation, combines with the boundary points, identifies and predicts the out-of-bounds behavior, outputs the out-of-bounds event sequence and transmits it to the signal output module.

2. The real-time monitoring and abnormal warning shooting training safety system according to claim 1, characterized in that, The training ground boundary point set includes a boundary point coordinate set, a boundary segment direction set, and a boundary contour structure. The action trajectory feature set includes a three-dimensional trajectory line, a velocity vector sequence, and a boundary approach marker. The posture recognition record includes an action label, a trajectory direction sequence, and a posture change pattern. The out-of-bounds event sequence includes a trajectory number, an out-of-bounds starting point, and offset duration information.

3. The real-time monitoring and abnormal warning shooting training safety system according to claim 1, characterized in that The boundary delineation module includes: A data acquisition sub-module that, through a millimeter-wave radar and an infrared positioning device, real-time obtains point cloud data in the training ground, performs data fusion by combining time synchronization and spatial registration, identifies the point cluster area where the change amplitude of the reflected signal intensity in each frame is less than the set floating interval within a continuous time period, filters the point cluster clusters located in the observation area based on the device position label, and generates the point cluster data volume for each frame; The set floating interval refers to setting an upper and lower limit range for the signal intensity change of the same spatial point in consecutive sampling frames; A coordinate calculation sub-module that, based on the point cluster data volume for each frame, selects multiple reference points with continuous positions in each frame, constructs a three-dimensional coordinate vector group according to the device positioning number, performs coordinate system conversion according to the ICP algorithm, calculates the spatial distance value between point pairs and the included angle value of three-point combinations, filters and eliminates the point groups that do not meet the continuous change interval of the included angle, and obtains the stable direction coordinate value interval; The ICP algorithm realizes the coordinate system conversion between point clusters and the screening of the stable direction interval by setting a point pair matching distance threshold, an error convergence criterion, and an initial positioning number; The continuous change interval of the included angle refers to the allowable fluctuation range of the change value of the vector included angle formed by multiple reference points between consecutive frames in three-dimensional space; A boundary generation sub-module that, according to the stable direction coordinate value interval, selects point segments with a direction change rate lower than the direction consistency threshold and performs structure combination, sorts the paths according to the direction consistency of the point segments, extracts the path vertices according to the Delaunay triangulation algorithm to construct a boundary structure, and generates the training ground boundary point set; The direction consistency threshold is set by statistically analyzing the distribution characteristics of the direction change rate in the direction stable coordinate interval; The Delaunay triangulation algorithm constructs a boundary vertex set based on the point segments with low direction change rate, and generates a continuous and consistent training ground boundary structure through circumcircle legality and boundary constraints.

4. The real-time monitoring and abnormal warning shooting training safety system according to claim 3, characterized in that, The position detection module includes: An attitude coordinate extraction sub-module that calls the three-dimensional coordinates of the boundary points in the training ground boundary point set, uses a millimeter-wave radar and an infrared positioning device to detect the presence state of a person within the boundary range, and real-time collects the human point cloud data, identifies the three-dimensional spatial coordinates of the person's shoulders, elbows, and wrists in each frame, and integrates them into a continuous frame coordinate sequence according to the time sequence to generate a three-dimensional attitude coordinate sequence; A speed trajectory calculation sub-module that calculates the distance between adjacent frames and the corresponding time interval based on the continuous coordinate points of the shoulders, elbows, and wrists in the three-dimensional attitude coordinate sequence, constructs a speed vector, and arranges it into a trajectory speed sequence according to the time sequence; An action feature generation sub-module that extracts the movement direction according to the speed change trend of each segment based on the trajectory speed sequence, analyzes the change of the direction vectors of the shoulders, elbows, and wrists in continuous frames, calculates the included angle offset range, identifies the continuously changing segments of the trajectory and arranges them into a behavior feature sequence, and outputs an action trajectory feature set.

5. The real-time monitoring and abnormal warning shooting training safety system according to claim 4, characterized in that, The action construction module includes: A trajectory extraction sub-module that obtains the position information of the shoulders, elbows, and wrists in the action trajectory feature set, arranges the trajectory data in frames according to the time sequence, filters out the signal continuous frame segments as the action interval, calculates the connection direction vector of the three points in each frame according to the coordinate change and normalizes it, establishes a trajectory vector set in each frame, and generates a part vector change sequence value; The signal continuous frame segment refers to a frame sequence in which the coordinate data of the three points of the shoulders, elbows, and wrists do not appear missing or interrupted in multiple adjacent frames in the time sequence; A direction analysis sub-module that calculates the vector included angle value between three points in adjacent frames based on the part vector change sequence value, obtains the change amplitude of the included angle between frames using cosine similarity, filters out the frame segments with an included angle change volatility lower than the direction stable threshold, extracts the continuously changing time segments within the frame segments, and obtains the direction stable trajectory interval; The specific formula for obtaining the change amplitude of the included angle between frames using cosine similarity is: ； Calculate the included angle perturbation fusion value; Among them, represents the angle perturbation fusion value of the i-th frame, represents the dot product value of the shoulder-elbow direction vector and the elbow-wrist direction vector of the i-th frame, represents the product of the moduli of the above two vectors, represents the displacement distance of the shoulder coordinates in the three-dimensional space from the i-th frame to the i+1-th frame, represents the difference in the included angle between the elbow-wrist vectors between the i-th frame and the i+1-th frame, is the displacement perturbation adjustment factor, is the trend adjustment factor, represents the number of frames with consistent forward continuous angle change trends, and i represents the current processed frame number in the time series; The included angle change volatility is obtained by calculating the standard deviation of the sequence of differences in vector included angles between adjacent frames; The direction stable threshold is obtained by collecting multiple stable action trajectories, calculating the standard deviation distribution of the included angle change volatility, and selecting the high quantile value as the threshold; An action recognition sub-module that analyzes the angle combination pattern between the vectors of the shoulders, elbows, and wrists in the trajectory segment according to the direction stable trajectory interval, identifies the action type of the person in the training ground, outputs the action type label, detects abnormal action types, and generates an attitude recognition record sequence.

6. The real-time monitoring and abnormal warning shooting training safety system according to claim 5, characterized in that, The abnormal recognition module includes: The moving trajectory extraction sub-module extracts the three-dimensional coordinates of the shoulders of the personnel in consecutive frames based on the pose recognition record, arranges the coordinate points in chronological order, identifies the spatial position change trend between consecutive coordinates, generates the moving trajectory of the shoulders in the training area, and generates a moving trajectory sequence; The path analysis sub-module arranges the coordinate points in chronological order according to the moving trajectory sequence, constructs a trajectory direction sequence based on the coordinate change trend, calculates the angular change between adjacent direction vectors, extracts the continuously changing angular trajectory segments, extends based on the end direction vector of the trajectory segment to generate a predicted moving path sequence, and generates a path offset feature segment set; The out-of-bounds prediction sub-module, according to the path offset feature segment set, combines the moving trajectory and the boundary point set, compares the spatial position relationship between the path and the boundary, filters out the path segments that overlap with the boundary, extracts the personnel numbers and time positions corresponding to the overlapping segments, and generates an out-of-bounds event sequence.

7. The real-time monitoring and abnormal warning shooting training safety system according to claim 6, characterized in that The specific formula for calculating the angular change between adjacent direction vectors is: ； Calculate the adjacent direction angle value; Among them, represents the direction angle value between the j-th frame and the (j + 1)-th frame, represents the dot product result of the direction vectors of the j-th frame and the (j + 1)-th frame, represents the adjustment coefficient of the direction difference of the j-th frame, represents the position offset between the j-th frame and the (j + 1)-th frame, represents the product of the magnitudes of the direction vectors of the j-th frame and the (j + 1)-th frame, represents the smoothing factor of the direction of the j-th frame, represents the index of the current frame.

8. The shooting training safety system for real-time monitoring and abnormal warning according to claim 6, characterized in that: The signal output module obtains the corresponding numbers and time positions in the out-of-bounds event sequence, locates the abnormal path using the boundary point cloud set, combines the recognition result of the abnormal action type, outputs the warning content, sends a warning prompt message and outputs the control instruction of the sound and light warning device, and generates an abnormal behavior response record; The control instruction of the sound and light warning device is to set different frequencies of sound and light alarms according to the length of the out-of-bounds path; The real-time abnormal behavior recording unit includes the personnel number, the abnormal area identifier, and the warning response method.

9. The real-time monitoring and abnormal warning shooting training safety system according to claim 8, characterized in that The signal output module includes: The event matching sub-module obtains the corresponding numbers and time positions in the out-of-bounds event sequence, matches the personnel identifier corresponding to the number, extracts the recognition result of the abnormal action type, arranges the behavior time sequence and action mark content of the number, and generates an event status synchronization value; The path positioning sub-module, according to the event status synchronization value, calls the spatial coordinate interval corresponding to the event time period in the boundary point cloud set, locates the spatial movement path of the personnel identifier within the corresponding time, filters out the coordinate point segments with boundary crossings in the path and extracts the continuous spatial displacement trajectory, and obtains the out-of-bounds path position sequence; The control instruction sub-module, based on the out-of-bounds path position sequence, combines the relationship between the event time and the trajectory number, extracts the corresponding warning level type information, matches the sound and light device control parameters according to the warning level type, and generates an abnormal behavior response record.

10. A shooting training safety method for real-time monitoring and abnormal warning, characterized in that, The method is used to implement the shooting training safety system for real-time monitoring and abnormal warning according to any one of claims 1-9, and the method includes: S1: Through the millimeter-wave radar and the infrared positioning device, real-time obtain the point cloud data in the training ground, extract the three-dimensional coordinates of multiple reference points, calculate the distance and direction angle between multiple reference points based on the point cloud spatial coordinates, classify the point segments with the same direction change amplitude, integrate them into a continuous boundary line segment path, and output it as the training ground boundary point set; S2: Obtain the set of boundary points of the training ground, detect the three-dimensional coordinate data of personnel within the training ground using point cloud data, obtain the spaces of the shoulders, elbows, and wrists, calculate the distance and time difference between adjacent time frames, generate velocity vectors, construct a continuous trajectory path of points in time series, and output it as an action trajectory feature set; S3: Obtain the action trajectory feature set, extract the change values of the direction angles between each pair of adjacent frames in the shoulder, elbow, and wrist trajectory vectors, extract the continuously changing trajectory segments of the direction, identify the action type and output the corresponding action label, and output it as a posture recognition record; S4: Obtain the posture recognition record, extract the shoulder coordinate sequence, identify the movement trajectory of the personnel, analyze the direction of speed change and the degree of continuous path deviation, combine with the position of the boundary points, identify and predict the out-of-bounds behavior, and output the out-of-bounds event sequence; S5: Obtain the corresponding numbers and time positions of the out-of-bounds event sequence, use the boundary point cloud set to locate the abnormal path, combine with the action label content in the corresponding posture recognition record, generate warning content according to the time index and the abnormal path number, send warning prompt information and output the control instructions for the sound and light equipment, and generate an abnormal behavior response record.

Images