Intelligent detection system and method for abnormal behavior in sports events

Abstract

The application provides a sports event abnormal behavior intelligent detection system and method, which identifies the relative position relationship between the feet of the front row players of both sides and the court markings, tracks the head position of the second receiving player of the other side from the moment of serving and hitting, and further estimates the corresponding line of sight direction, and synchronously tracks the real-time positions of all attack candidate players of the front row of the own side, determines an attack reference point based on the line of sight direction and the real-time positions of all attack candidate players, and further constructs a dynamic space connection line pointing to the attack reference point, and further monitors the movement of the non-attack core players of the front row of the own side, and further determines the continuous shielding time length of the non-attack core players of the front row of the own side shielding the dynamic space connection line through three-dimensional space analysis, and if the continuous shielding time length exceeds a preset threshold, it is determined that a shielding station violation occurs. The scheme of the application can realize effective identification of hidden and combined violation behaviors.

Description

Technical Field

[0001] This application relates to the field of intelligent detection technology for abnormal behavior, and more specifically, to an intelligent detection system and method for abnormal behavior in sports events. Background Technology

[0002] Intelligent detection of abnormal behavior refers to a method that uses a series of computational processes, such as multimodal sensor network acquisition, time-series dynamic data parsing, implicit pattern modeling, and abnormal pattern matching, to transform the continuous behavioral flow of individuals or groups in real-world scenarios into a structured representation with semantic annotation and risk quantification, thereby enabling automatic identification and early warning of target behavior shifts from normal baselines to abnormal deviations.

[0003] In existing intelligent detection processes for abnormal behavior in sports events, traditional methods often rely on the separate analysis of athletes' static positions or simple dynamic trajectories. This makes it difficult to effectively identify and quantify concealed violations that require the simultaneous integration of dynamic line of sight, three-dimensional spatial relationships, and temporal behavior. For example, in volleyball, for covert positioning violations, traditional methods may only be able to detect positional errors at the moment of serving or track athletes' movements in a two-dimensional plane. However, they cannot fully capture the spatiotemporal behavioral chain of non-offensive core players dynamically moving based on positional errors to continuously obstruct the opponent's setter's view of our offensive core. This makes it impossible to quantify dynamic three-dimensional spatial obstruction into objective indicators corresponding to the rules, ultimately leading to a break in the violation chain and resulting in misjudgments or omissions based on partial data in the rulings. Therefore, how to effectively identify concealed and combined violations has become a challenge for the industry. Summary of the Invention

[0004] This application provides an intelligent detection system and method for abnormal behavior in sports events, which can effectively identify covert and combined violations.

[0005] Firstly, this application provides an intelligent detection method for volleyball positioning violations, applied to an intelligent detection system for abnormal behavior in sports events. The method includes the following steps: Just before the serving player hits the ball, acquire video images of the court and identify the relative positions of the feet of the front-row players from both sides to the court markings; Starting from the moment the ball is hit, the head position of the opposing setter is tracked based on the video images of the court to estimate the corresponding line of sight, and the real-time position of all attacking candidates in the front row of the team is tracked simultaneously. Based on the line of sight and the real-time positions of all attacking candidates, the attacking candidates within the extended area of ​​the line of sight are identified as the observed attacking reference points, thereby constructing a dynamic spatial connection from the setter's head to the attacking reference points. Based on the dynamic spatial connection, the movement of the team's front row non-offensive core players is monitored, and then the duration of continuous occlusion of the team's front row non-offensive core players' body envelope by the dynamic spatial connection is determined through three-dimensional spatial analysis. If the duration of continuous obstruction exceeds a preset threshold, and the player is judged to have a positional error as defined by the volleyball rules based on the relative positional relationship, then a violation of the cover position rule is determined to have occurred.

[0006] In some embodiments, starting from the moment the ball is struck during the serve, tracking the head position of the opposing setter based on the court video image and then estimating the corresponding line of sight specifically includes: Starting from the moment the ball is served, the video images of the court are continuously analyzed frame by frame to maintain cross-frame identity tracking of the opposing setter. In each video frame, the head region of the opposing setter is detected and tracked to locate facial key points for posture estimation. Based on the image coordinates of the facial key points and their corresponding 3D head model, the head rotation posture is calculated, and then the corresponding gaze direction is estimated.

[0007] In some embodiments, synchronously tracking the real-time positions of all attacking candidates in the front row of one's own team specifically includes: Based on the athlete's initial identity associated with the relative positional relationship, multi-target tracking is initiated for the team's front-row players in the field video image; In the video images of the arena, the identification of each player in the front row of their team is maintained by associating the data of the detection box with the trajectory. For each member of the team whose identity is successfully maintained, the real-time position of all attacking candidates in the team's front row is tracked through the world coordinate system mapping relationship.

[0008] In some embodiments, based on the line of sight and the real-time positions of all attacking candidate players, attacking candidate players within the extended area of ​​the line of sight are determined as observed attacking reference points, thereby constructing a dynamic spatial connection from the setter's head to the attacking reference points. Specifically, this includes: Based on the line of sight and the position of the setter's head in the world coordinate system, a regular spatial region is constructed with the position of the setter's head as the vertex and extending along the line of sight. The set of attacking candidate players located within the rule space area is obtained by filtering based on the real-time positions of all attacking candidate players. Based on preset selection rules, a unique player is selected from the set of attacking candidate players as the attacking reference point to be observed; Based on the real-time position coordinates of the setter's head and the attack reference point, a dynamic spatial connection is constructed from the setter's head to the attack reference point.

[0009] In some embodiments, monitoring the movement of the team's front-row non-offensive core players based on the dynamic spatial connection specifically includes: Based on the dynamic spatial connection, the non-attack core players that need to be monitored are extracted from all the attacking candidate players in the front row of our side. Continuously acquire the real-time location coordinates of each non-offensive core player to monitor their movement trajectory.

[0010] In some embodiments, determining the continuous occlusion duration of the dynamic spatial connection line by the body envelope of the team's front-row non-offensive core player through three-dimensional spatial analysis specifically includes: A three-dimensional geometric model of the body envelope of each non-offensive core player is constructed based on the real-time position and posture of each player. By connecting the three-dimensional geometric model with the dynamic space, a three-dimensional occlusion determination is performed to obtain a binary state of whether occlusion has occurred at different time points; The duration of continuous occlusion of the dynamic spatial connection line by the body envelope of the non-offensive core player in the front row is determined based on the binary state within a continuous time period.

[0011] In some embodiments, if the duration of continuous obstruction exceeds a preset threshold, and the player is determined to have committed a positional error as defined by the volleyball rules based on the relative positional relationship, then determining that a covert positioning violation has occurred specifically includes: The duration of continuous occlusion is compared with a preset threshold to determine whether the timeout condition is met. Based on the relative positional relationship, determine whether the non-offensive core player who caused the continuous blocking duration constitutes a positional error as defined by the volleyball rules; If both the timeout condition and the position error condition are met simultaneously, it is determined that the non-offensive core player has violated the cover positioning rule.

[0012] Secondly, this application provides an intelligent detection system for abnormal behavior in sports events, including an intelligent detection unit for position violations, wherein the intelligent detection unit for position violations includes: The acquisition module is used to acquire video images of the court just before the serving player hits the ball, and to identify the relative positional relationship between the feet of the front row players of both sides and the court markings. The processing module is used to track the head position of the opposing setter based on the video image of the court from the moment of the serve and hit the ball, and then estimate the corresponding line of sight, and simultaneously track the real-time position of all attacking candidates in the front row of the team. The processing module is also used to determine the attacking candidate players within the extended area of ​​the line of sight as the observed attacking reference points based on the line of sight and the real-time positions of all attacking candidate players, thereby constructing a dynamic spatial connection line from the setter's head to the attacking reference points. The processing module is also used to monitor the movement of the team's front row non-offensive core players based on the dynamic spatial connection, and then determine the duration of continuous occlusion of the dynamic spatial connection by the body envelope of the team's front row non-offensive core players through three-dimensional spatial analysis. The execution module is used to determine that a cover position violation has occurred if the duration of continuous obstruction exceeds a preset threshold and the player's position is determined to be incorrect according to the relative positional relationship as defined by the volleyball rules.

[0013] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described intelligent detection method for volleyball positioning violations.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described intelligent detection method for volleyball positioning violations.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The intelligent detection system and method for abnormal behavior in sports events provided in this application firstly determines the observed offensive reference point by tracking the head and gaze direction of the opposing setter and combining it with the real-time position of the team's attacking candidate players. Then, a dynamic spatial line is constructed from the setter's head to this reference point. This process transforms the subjective and vague "gaze direction" rule into a geometric line segment in three-dimensional space with a clear starting point, direction, and target endpoint, which can be calculated and tracked in real time. This objectifies the visual behavior of "observation" into a quantifiable mathematical baseline for subsequent spatial analysis, providing a unique and continuous spatial reference for objectively judging "obstruction" behavior. Subsequently, through three-dimensional spatial analysis, the occlusion status of the body envelope of a specific non-offensive core player on the aforementioned dynamic spatial line is continuously monitored, and the duration of continuous occlusion is calculated. This process transforms the qualitative judgment of "whether the gaze is obstructed" into a quantifiable mathematical baseline for "whether the three-dimensional occlusion of a specific object on a specific spatial line segment continuously meets a specific time threshold." The precise measurement of temporal physical indicators objectifies the referee's experience of "not penalizing brief, accidental obstructions" into a repeatable and verifiable technical criterion based on a preset threshold, providing a direct quantitative tool for identifying intentional, non-accidental obstruction behavior. Furthermore, by combining the dynamic indicator of continuous obstruction duration exceeding a preset threshold with the static violation criterion of positional error determined based on the relationship between the foot and the line just before the serve, a temporal correlation and logical AND judgment are performed after the unified time origin of "serving and hitting the ball." This process can bind the two independent events of "positional error" and "dynamic movement to continuously obstruct vision," which are detected separately in traditional methods, into a coherent chain of violations with a clear sequential logic (first incorrect positioning, then intentional obstruction) and a common purpose (covering the attack) in time and space. This technically achieves the automated reconstruction and identification of the complete "covering positioning violation" described by the rules. In summary, this scheme can effectively identify concealed and combined violations. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an intelligent detection method for volleyball positioning violations according to some embodiments of this application; Figure 2 This is a flowchart illustrating the process of determining the line of sight direction according to some embodiments of this application; Figure 3 This is a flowchart illustrating the determination of timeout conditions according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of the intelligent detection unit for station violations shown in some embodiments of this application; Figure 5This is an internal structural diagram of a computer device that implements an intelligent detection method for volleyball positioning violations, according to some embodiments of this application. Detailed Implementation

[0017] To better understand the technical solutions in this embodiment, the technical solutions in this embodiment will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0018] refer to Figure 1 The figure is a flowchart illustrating an intelligent detection method for volleyball positioning violations according to some embodiments of this application. The intelligent detection method for volleyball positioning violations mainly includes the following steps: In step 101, at the moment before the serving player hits the ball, a video image of the court is acquired, and the relative positional relationship between the feet of the front row players of both sides and the court markings is identified.

[0019] In practice, one or more calibrated high-definition cameras deployed at fixed locations on the field can continuously capture a video stream covering the entire field at a preset frame rate. A reference image frame is precisely determined and captured the instant before the serving player hits the ball by a trigger mechanism based on motion recognition. The video stream is transmitted to the processing unit in real time via a network. In the processing unit, the reference image is first preprocessed with distortion correction and color enhancement. Then, a target detection model based on a convolutional neural network is used to segment the athletes in the image, and a key point detection model is used to locate the pixel coordinates of the foot key points within each segmented region. Simultaneously, an algorithm based on Hough transform or semantic segmentation is used to extract the pixel-level contours of the field markings. Then, using the known dimensions of the field as an absolute scale constraint, a mapping relationship is established from the two-dimensional pixel coordinate system of the image to the three-dimensional world coordinate system of the field by solving the homography matrix. Based on this mapping, the pixel coordinates of the foot key points and the field marking contours are uniformly transformed to the world coordinate system to obtain quantified spatial positions. Finally, the relative positional relationship is identified and recorded by calculating the geometric distance between the player's foot position and relevant markings (such as the center line) and comparing it with a rule threshold.

[0020] It should be noted that the relative positional relationship mentioned in this application refers to geometric parameters (such as distance values) obtained through quantitative calculation under a unified world coordinate system, used to determine whether the athlete's initial stance is compliant. Its physical significance lies in providing a traceable objective benchmark for subsequent violation judgments. The video image of the competition field refers to a two-dimensional image containing known spatial reference objects, captured by a calibrated fixed-view camera, which is the data basis for subsequent spatial geometric calculations. The instant before the serving player hits the ball refers to the unique instant defined according to the volleyball rules, used to determine the legality of the athlete's initial stance, which is accurately captured by technical means. The target detection model, key point detection model, and datum line extraction algorithm are all well-known technical means in the field, and this application does not limit their specific network structure and implementation variations.

[0021] In step 102, starting from the moment the ball is hit, the head position of the opposing setter is tracked based on the video image of the court to estimate the corresponding line of sight, and the real-time position of all attacking candidates in the front row of our side is tracked simultaneously.

[0022] In some embodiments, reference Figure 2 As shown in the figure, this is a flowchart illustrating the process of determining the direction of gaze in some embodiments of this application. Starting from the moment the ball is struck during the serve, the direction of gaze can be estimated by tracking the head position of the opposing setter based on the video image of the court. This can be achieved through the following steps: First, in step 1021, starting from the moment the ball is hit, the video image of the court is continuously analyzed frame by frame to maintain cross-frame identity tracking of the opposing setter. Then, in step 1022, in each video frame, the head region of the tracked and determined opposing setter is detected, thereby locating facial key points for posture estimation. Finally, in step 1023, the head rotation posture is calculated based on the image coordinates of the facial key points and their corresponding three-dimensional head model, thereby estimating the corresponding gaze direction.

[0023] In specific implementation, starting from the moment the ball is served, the video images of the court are continuously analyzed frame by frame. Maintaining cross-frame tracking of the opposing setter's identity can be achieved in the following way: For example, based on the identities of the front-row players of both sides and their initial positions in the world coordinate system, a multi-target tracking process is initiated after the moment the ball is served. This process first uses the target detection model to detect athletes in each frame of the subsequent continuously input video image frame sequence, obtaining their bounding boxes. Then, a multi-target tracking algorithm based on data association is used to match and associate all athlete bounding boxes detected in the current frame with existing target trajectories. To accurately maintain tracking of the opposing setter, the matching and association process not only calculates the image intersection-union ratio between detection boxes, but more importantly, applies the established world coordinate system mapping relationship to each detected target, setting its baseline... The center of the foot or detected key points are mapped to the world coordinate system to obtain its current estimated world plane position. Then, the Euclidean distance between this position and the predicted world plane positions of each existing trajectory at the previous moment is calculated. The consistency of this spatial distance is used as one of the core association costs, combined with the appearance feature similarity cost, to perform global optimal matching. For detection boxes that are successfully associated with the specific identity of the opposing setter, the trajectory corresponding to that identity is updated, thereby realizing cross-frame identity tracking of the opposing setter. If a brief occlusion causes the association to fail, the Kalman filter prediction algorithm is used to make a short-term prediction of its position in the world coordinate system to maintain the continuity of the trajectory until it is re-detected and associated. As a preferred embodiment, in the matching association process, the Hungarian algorithm or the greedy matching algorithm can be used to solve the global optimal matching to balance tracking accuracy and computational efficiency.

[0024] It should be noted that the cross-frame identity tracking described in this application refers to generating a temporally continuous and identity-consistent motion trajectory for a specific athlete target (such as the opposing setter) by maintaining the identity identifier of the target in a continuous sequence of video frames.

[0025] In specific implementation, detecting and tracking the head region of the opposing setter in each video frame, and then locating facial key points for pose estimation, can be achieved in the following way: For example, for the image region representing the opposing setter determined through the aforementioned cross-frame identity tracking, head region detection is first performed in the upper half of the bounding box; the head region detection can be completed using a lightweight head detector, or the estimated range of the head region can be directly derived from the coordinates of the shoulder and neck key points based on the human skeleton key points already provided by the target detection model; after locating the head region, a facial key point detection model is further applied within this region; the facial key point detection model is configured to output... A series of predefined two-dimensional image pixel coordinates of facial feature points, including at least the tip of the nose, the inner corners of the left and right eyes, and the corners of the left and right mouth; to cope with facial blurring or partial occlusion that may occur during motion, the detection process can incorporate temporal smoothing constraints, that is, by utilizing the continuity of facial key point positions between consecutive frames, the detection results of the current frame are smoothed through a filtering algorithm, thereby obtaining stable facial key point coordinates for pose estimation; as a preferred embodiment, the temporal smoothing constraints can be implemented by a first-order or second-order Kalman filter to achieve a balance between tracking accuracy and response speed; in other embodiments, a key point trajectory prediction method based on a recurrent neural network can also be used, which is not limited in this application.

[0026] It should be noted that the facial key points mentioned in this application refer to a set of two-dimensional image coordinates of a series of feature points (such as the tip of the nose, the corner of the eye, etc.) with fixed anatomical significance obtained from the head image region.

[0027] In specific implementation, the head rotation posture is calculated based on the image coordinates of the facial key points and their corresponding 3D head model, and the corresponding gaze direction is estimated. This can be achieved in the following way: First, a standardized 3D head model is defined, which consists of 3D coordinate points that correspond one-to-one with the point set output by the facial key point detection model. Their relative positions are predefined based on the average facial anatomy. Next, a set of 2D facial key point image coordinates of the opposing setter detected in the current frame are paired with the corresponding 3D points in the standardized 3D head model. A rotation matrix and translation vector are solved using the perspective n-point algorithm, so that the 3D head model points are projected onto the image plane after rotation and translation, and then compared with the detected... The reprojection error between two-dimensional keypoints is minimized; the obtained rotation matrix represents the three-dimensional rotational posture of the head relative to the camera; then, the vector representing the head's forward orientation is extracted from the rotation matrix and transformed to the camera coordinate system through the matrix to obtain the preliminary three-dimensional vector of the head orientation in this frame; finally, combined with the known extrinsic parameters of the camera relative to the world coordinate system, this head orientation vector is transformed from the camera coordinate system to the world coordinate system and normalized to estimate the gaze direction of the opposing setter in the current moment, defined in the world coordinate system; in other embodiments, an end-to-end gaze estimation algorithm based on a deep learning network can also be directly used, and this application does not limit it to this.

[0028] It should be noted that the gaze direction mentioned in this application refers to a three-dimensional unit vector defined in the world coordinate system, obtained by estimating head posture, used to characterize the direction in space of the front of the opponent's setter's head (or the visual attention after fine-tuning).

[0029] In some embodiments, the real-time positions of all attacking candidates in the front row of one's own team can be simultaneously tracked using the following steps: Based on the athlete's initial identity associated with the relative positional relationship, multi-target tracking is initiated for the team's front-row players in the field video image; In the video images of the arena, the identification of each player in the front row of their team is maintained by associating the data of the detection box with the trajectory. For each member of the team whose identity is successfully maintained, the real-time position of all attacking candidates in the team's front row is tracked through the world coordinate system mapping relationship.

[0030] In specific implementation, based on the initial identities of the athletes associated with the relative positional relationships, multi-target tracking of the team's front-row players in the match video images can be initiated in the following way: for example, using the identities of both teams' front-row players recorded in the relative positional relationships and their initial positions in the world coordinate system as a reference, after the serve and hit are triggered, an independent multi-target tracker is initialized for the team's front-row players; in subsequent processing, for the continuously input match video image frame sequence, this tracker first uses the target detection model to detect the athletes in each frame, obtaining the boundaries of all athletes. The tracker first identifies the target player in the front row. Then, based on the initial identity information, it filters out high-confidence target boxes belonging to the category of "team's front row player" from all the target boxes. The filtering can be based on the coarse classification probability provided by the detection model, or combined with the initial position information (e.g., the area located in the team's half of the court and close to the net) for logical judgment. For the selected target boxes of the team's front row player, the tracker creates an initial trajectory for each of them. The trajectory records the target player's identity, initial image bounding box, and initial world coordinates obtained by the world coordinate system mapping relationship, thereby completing the multi-target tracking of the team's front row player.

[0031] It should be noted that the multi-target tracking described in this application refers to the technical process of simultaneously performing cross-frame detection, association, and trajectory prediction of multiple athlete targets in a video.

[0032] In specific implementation, maintaining the identity of each team's front-row player by associating detection boxes with trajectories in the video image of the competition can be achieved in the following way: For example, after the multi-target tracking is started, for each newly input image frame, the tracker performs a data association step to maintain the continuity of identity identification; this step first obtains all "team's front-row player" detection boxes obtained by the target detection model in the current frame, and uses the world coordinate system mapping relationship to map the bottom center point of each detection box to the world coordinate system to obtain its corresponding world plane position estimate; then, it calculates the Euclidean distance between the world position of each detection box in the current frame and the predicted world position of all existing trajectories in the tracker at the previous moment, forming a distance cost matrix; at the same time, it can calculate the detection boxes and trajectories in the image feature space (such as the appearance feature vector extracted by the convolutional neural network). The similarity on the distance cost and appearance cost is used to form an appearance cost matrix. A comprehensive association cost matrix is ​​obtained by combining the distance cost and appearance cost (e.g., weighted summation). Then, a data association algorithm (such as the Hungarian algorithm) is used to solve the comprehensive cost matrix to assign a detection box in the current frame to each existing trajectory, or mark it as unmatched. Successfully matched detection boxes are used to update the state of the corresponding trajectory (including its image bounding box, world coordinates, and appearance features) to maintain the identity of the trajectory. For unmatched trajectories, a motion prediction model (such as Kalman filtering based on the constant velocity assumption) is used to predict its world position and image position in the current frame to keep the trajectory alive and deal with short-term occlusion or detection failure. Unmatched detection boxes may be initialized as new trajectories, but in this step, the main purpose is to maintain the identity of known front-row players rather than introduce new targets.

[0033] In practice, for each front-row player whose identity has been successfully maintained, the real-time positions of all attacking candidates in the front row can be tracked using a world coordinate system mapping relationship. This can be achieved as follows: For each front-row player's trajectory that has successfully maintained its identity through the aforementioned data association steps, the tracker outputs its latest image bounding box in each video frame. Using the world coordinate system mapping relationship, the bottom center point of this bounding box (or more precisely, the player's foot keypoint located using the keypoint detection model) is transformed from image pixel coordinates to the world coordinate system, thus obtaining the player's two-dimensional world plane coordinates (X, Y) in the current frame. This coordinate represents the player's real-time position. By continuously performing this process on all successfully maintained identities, all tracked front-row players can be obtained. The real-time position set is defined as follows: "Attack candidate players" are defined in this application as all players in the frontcourt area (e.g., within and near the three-meter line). Therefore, after obtaining the real-time position set, the known equations of the court markings in the world coordinate system (e.g., the position of the three-meter line) can be used to determine whether each player is located in the frontcourt area, thereby filtering or marking all "attack candidate players" and their corresponding real-time positions, completing the tracking of "the real-time positions of all attack candidate players in the frontcourt of this team." As a preferred embodiment, the determination of the frontcourt area can be achieved by comparing the player's Y-coordinate (assuming the Y-axis is perpendicular to the net) with the Y-coordinate threshold corresponding to the three-meter line. In other embodiments, the range of "attack candidate players" can also be dynamically defined according to the tactical stage of the game; this application does not limit this.

[0034] It should be noted that the world coordinate system mapping relationship described in this application refers to the mathematical transformation relationship established from the two-dimensional pixel coordinate system of the camera image to the three-dimensional world coordinate system based on the volleyball court. Its physical significance lies in uniformly transforming the pixel positions in the image to a real spatial coordinate system with absolute physical dimensions and a unified origin, so that the calculation of all athletes' positions, line of sight, etc., has measurability and spatial consistency.

[0035] In step 103, based on the line of sight and the real-time positions of all attacking candidate players, the attacking candidate players within the extended area of ​​the line of sight are determined as the observed attacking reference points, thereby constructing a dynamic spatial connection from the setter's head to the attacking reference points.

[0036] In some embodiments, based on the line of sight and the real-time positions of all attacking candidates, the attacking candidates within the extended area of ​​the line of sight are determined as the observed attacking reference points, and a dynamic spatial connection from the setter's head to the attacking reference points is constructed. This can be achieved through the following steps: Based on the line of sight and the position of the setter's head in the world coordinate system, a regular spatial region is constructed with the position of the setter's head as the vertex and extending along the line of sight. The set of attacking candidate players located within the rule space area is obtained by filtering based on the real-time positions of all attacking candidate players; Based on preset selection rules, a unique player is selected from the set of attacking candidate players as the attacking reference point to be observed; Based on the real-time position coordinates of the setter's head and the attack reference point, a dynamic spatial connection is constructed from the setter's head to the attack reference point.

[0037] In specific implementation, based on the line of sight and the position of the setter's head in the world coordinate system, a regular spatial region extending along the line of sight with the setter's head position as the vertex can be constructed in the following way: for example, using the estimated line of sight unit vector defined in the world coordinate system as the central axis, and the real-time world coordinates of the setter's head as the vertex; based on this, a spatial cone is constructed as the regular spatial region; the central axis of this cone coincides with the line of sight vector, and its half-angle is a preset adjustable parameter used to define the width range of the line of sight extension; the half-angle... The value of the angle parameter can be preset based on statistical analysis of the observation habits of setters in volleyball matches. A larger value can cover a wider potential observation range, while a smaller value requires higher accuracy in estimating the direction of sight. The vertex, central axis, and half-angle completely define a geometric space range for determining whether a player is in the observed area. As a preferred embodiment, the regular spatial region adopts the spatial cone model because it can better simulate the spatial distribution characteristics of the human eye's field of vision. In other embodiments, other regular geometric shapes such as pyramids and sector prisms can also be used for approximation, and this application does not limit this.

[0038] It should be noted that the regular spatial area mentioned in this application refers to the three-dimensional spatial range defined by a preset geometric shape (such as a cone) extending along the line of sight of the setter, with the position of the setter's head as the vertex.

[0039] In specific implementation, the set of attack candidate players located within the regular spatial area can be obtained by filtering based on the real-time positions of all attack candidate players. For example: obtain the real-time positions of all attack candidate players in the front row of the team obtained through continuous tracking; for each attack candidate player, calculate the relative geometric relationship between its position and the regular spatial area to determine whether it is located within the area; for the cone model, the determination can be achieved sequentially through the following calculations: first, calculate the vector pointing from the vertex of the cone to the position of the attack candidate player; then, calculate the angle between the vector and the central axis of the cone; then, determine whether the angle is less than or equal to half an angle of the cone; if it is less than or equal to, determine that the player is located within the regular spatial area and add it to the candidate set; traverse the real-time positions of all attack candidate players and repeat the above determination process to finally obtain all attack candidate players located within the extended area of ​​the line of sight, forming the set of attack candidate players.

[0040] It should be noted that the set of attacking candidate players mentioned in this application refers to the set of all front-row players of our team who are determined to be within the possible observation range of the opposing setter at a specific moment, based on their real-time position and the geometric relationship between their position and the rule space area.

[0041] In specific implementation, determining a single player from the set of attack candidate players as the observed attack reference point according to a preset selection rule can be achieved in the following ways: For example, after obtaining the set of attack candidate players, apply the preset selection rule based on the number of players in the set; if the set is empty, it is determined that there is no clear attack reference point at the current moment, and the reference point from the previous moment can be used or subsequent frame data can be waited for; if there is only one player in the set, that player is directly determined as the observed attack reference point; if there are multiple players in the set, selection is made according to a preset geometric priority rule; the geometric priority rule can be defined as: selecting its position relative to the circle... The player with the smallest included angle along the cone's central axis is used as the attack reference point. This rule corresponds to selecting the player closest to the setter's line of sight. In another preferred embodiment, the player closest to the cone's apex can also be selected as the attack reference point. This selection rule needs to be predefined and configured in the system. Its purpose is to determine the most likely target to be observed from multiple candidates, thereby giving the subsequent occlusion analysis a clear direction. As a preferred embodiment, the minimum included angle rule is preferred. In other embodiments, the player's movement trend and historical role information can also be combined for comprehensive decision-making. This application does not limit the specific variations of the selection rule.

[0042] It should be noted that the attack reference point mentioned in this application refers to a unique front-row player of our team determined from the set of attack candidate players according to preset rules, used to characterize the specific attack target that is most likely to be observed by the opponent's setter at the current moment.

[0043] In specific implementation, based on the real-time position coordinates of the setter's head and the attack reference point, constructing a dynamic spatial connection from the setter's head to the attack reference point can be achieved in the following way: after determining the attack reference point, the world coordinates of the reference point obtained through the tracking are acquired in real time, and the world coordinates of the setter's head are also acquired. Mathematically, the dynamic spatial connection is a spatial vector, which is calculated by subtracting the world coordinate vector of the setter's head from the world coordinate vector of the attack reference point. The resulting vector is the line connecting the head to the reference point. The direction of this line vector represents the direction from the setter's line of sight to their observed target, and its length is the spatial distance between the two. Since the positions of the setter's head and the attack reference point are updated in real time, the above coordinate acquisition and vector calculation process needs to be repeated every frame to construct a spatial connection that dynamically updates over time, i.e., the dynamic spatial connection.

[0044] It should be noted that the dynamic spatial connection described in this application refers to a spatial vector line segment pointing from the head position of the setter to the position of the attack reference point in a three-dimensional world coordinate system. Its physical meaning is to geometrically represent the direct visual connection between the setter and the observed target in real time.

[0045] In step 104, the movement of the non-offensive core players in the front row of the team is monitored based on the dynamic spatial connection, and then the duration of continuous occlusion of the dynamic spatial connection by the body envelope of the non-offensive core players in the front row is determined by three-dimensional spatial analysis.

[0046] In some embodiments, monitoring the movement of one's own front-row non-offensive core players based on the dynamic spatial connection can be achieved through the following steps: Based on the dynamic spatial connection, the non-attack core players that need to be monitored are extracted from all the attacking candidate players in the front row of our side. Continuously acquire the real-time location coordinates of each non-offensive core player to monitor their movement trajectory.

[0047] In specific implementation, the non-attack core player to be monitored, extracted from all attacking candidate players in the front row based on the dynamic spatial connection, can be achieved in the following way: According to the definition of the dynamic spatial connection, it connects the setter's head to the observed attack reference point; thus, the attack reference point, as the specific manifestation of the setter's visual focus, is defined as the "attack core player" in this step; based on this definition, the non-attack core player to be monitored is the player remaining after excluding the attack reference point (i.e., the attack core player) from the set of all attacking candidate players in the front row. The extraction process specifically involves: at each processing moment, obtaining the identity identifier of the attack reference point at the current moment, and simultaneously obtaining a list of identity identifiers of all attack candidate players in the front row of the team; by comparing the identity identifiers, marking and identifying all players in the list whose identity identifiers differ from those of the attack reference point as non-attack core players to be monitored at the current moment; as a preferred embodiment, to ensure the continuity of the monitoring target, when the attack reference point cannot be updated temporarily due to tracking loss or other reasons, the identity of the attack reference point at the previous valid moment can be used to extract the non-attack core players.

[0048] It should be noted that the non-core offensive players mentioned in this application refer to all other players in the front row of the team, except those who are identified as the offensive reference point (i.e., the core offensive players). Their physical meaning is that they are the set of target objects that may carry out cover-like positioning behavior and therefore need to be closely monitored for their spatial relationship with the dynamic space connection.

[0049] In specific implementation, continuously acquiring the real-time position coordinates of each non-offensive core player to monitor their movement trajectory can be achieved in the following way: After completing the identification of the non-offensive core player, the system queries and reads the tracking trajectory corresponding to these identifications from the multi-target trackers of the front-line players based on their identifications; for each identification marked as a non-offensive core player, the system accesses its corresponding trajectory and extracts the latest world coordinates from it, which is its real-time position in the current frame; by repeating this query and extraction operation in each video frame or processing cycle, the system can obtain a series of position points of each non-offensive core player over time, thereby forming their movement trajectory; as a preferred embodiment, in order to smooth sensor noise and detect jitter, the trajectory data can be low-pass filtered after extracting the real-time position coordinates, but this application does not limit whether such post-processing is performed.

[0050] It should be noted that the movement trajectory mentioned in this application refers to a series of position coordinates in the world coordinate system that are arranged in chronological order for each non-offensive core player, generated by the multi-target tracking technology.

[0051] In some embodiments, determining the duration of continuous occlusion of the dynamic spatial connection line by the body envelope of a non-offensive core player in the front row through three-dimensional spatial analysis can be achieved by the following steps: A three-dimensional geometric model of the body envelope of each non-offensive core player is constructed based on the real-time position and posture of each player. By connecting the three-dimensional geometric model with the dynamic space, a three-dimensional occlusion determination is performed to obtain a binary state of whether occlusion has occurred at different time points; The duration of continuous occlusion of the dynamic spatial connection line by the body envelope of the non-offensive core player in the front row is determined based on the binary state within a continuous time period.

[0052] In specific implementation, constructing a three-dimensional geometric model of the body envelope based on the real-time position and posture of each non-offensive core player can be achieved in the following way: First, for each monitored non-offensive core player, obtain their real-time position coordinates (X, Y in the world coordinate system) and a set of two-dimensional image coordinates of key human skeletal points obtained through the human posture estimation model; then, using the world coordinate system mapping relationship and a depth estimation algorithm based on monocular vision (or the known multi-camera triangulation principle), the two-dimensional human key points are elevated to three-dimensional space to obtain the three-dimensional coordinates of these key points in the world coordinate system; wherein, the depth estimation algorithm can solve for proportional constraints based on prior statistical values ​​of the bone length of various parts of the human body; then, a simplified human geometric model is used to represent the body envelope, for example, making Multiple cylinders are used to approximate major limb parts such as the torso, upper arm, forearm, thigh, and lower leg. The spatial position and orientation of each cylinder are determined by the three-dimensional coordinates of key points at its two ends (e.g., the torso cylinder is defined by the key points at the shoulder and hip). Its radius is set to a preset parameter related to the athlete's body shape based on prior anthropometry data. In this way, a model is constructed for each non-offensive core player at the current moment, consisting of multiple simple three-dimensional geometric bodies (cylinders), which can represent the approximate area occupied by their body in space, i.e., the three-dimensional geometric model of the body envelope. As a preferred embodiment, the multi-cylinder model can be used because it achieves a good balance between computational complexity and geometric fit. In other embodiments, elliptical bodies, cuboids, or more refined skinned mesh models can also be used for approximation, and this application does not limit this.

[0053] It should be noted that the three-dimensional geometric model described in this application refers to an approximate model constructed based on the athlete's posture key points and composed of simple geometric shapes (such as cylinders) to characterize the range occupied by the athlete's body in space.

[0054] In specific implementation, the three-dimensional occlusion determination is performed by connecting the three-dimensional geometric model with the dynamic space to obtain the binary state of whether occlusion has occurred at different time points. This can be achieved in the following way: After obtaining the three-dimensional geometric model (e.g., multiple cylinders) of the body envelope of a non-offensive core player at a specific moment and the dynamic space connection at the same moment (defined as a three-dimensional space line segment connecting the setter's head point P_h and the offensive reference point P_r), geometric intersection detection in three-dimensional space is performed. The determination is performed on each constituent geometry (e.g., each cylinder) in the player's body envelope. For each cylinder, the core is to determine whether the space line segment intersects with the cylinder. This can be achieved by calculating the spatial relationship between the line segment and the central axis of the cylinder, for example, calculating the line segment closest to the axis of the cylinder. If a point is located between the two endpoints of a line segment and its distance to the axis is less than the radius of the cylinder, then the cylinder is determined to intersect the line segment. If any of the constituent geometric bodies of a player is determined to intersect the dynamic space connection, then at that moment, the player's body envelope constitutes an occlusion of the dynamic space connection, and an occlusion state value of "1" (or "true") is generated for this time point. Conversely, if none of the constituent geometric bodies intersect the connection, an occlusion state value of "0" (or "false") is generated. For each monitored non-offensive core player, the above determination process is repeated at each processing time point (such as the timestamp corresponding to each video frame), thereby generating a sequence of "0" and "1" arranged in chronological order for each player, which is the binary state sequence representing whether occlusion occurs at different time points.

[0055] It should be noted that the binary state mentioned in this application refers to the judgment result (usually represented by "1" or "0") obtained by three-dimensional spatial analysis at a specific point in time, which characterizes whether the body envelope of a non-offensive core player constitutes occlusion of the dynamic spatial connection. Its physical meaning is to record the existence or non-existence of spatial occlusion relationship at each sampling moment in the simplest form.

[0056] In specific implementation, determining the continuous occlusion duration of the dynamic spatial connection line continuously obscured by the body envelope of the team's front-row non-offensive core player based on the binary states within a continuous time period can be achieved in the following way: For example, for a specific non-offensive core player, obtain the binary state sequence generated by him within the continuous time period (e.g., from the time after the serve to the time before the attack occurs); each element in the sequence corresponds to a specific sampling time point, either equally spaced or unequally spaced; analyze the sequence to find all consecutive segments with a state value of "1"; for each such consecutive "1" segment, the corresponding continuous occlusion duration is calculated as follows: subtract the timestamp corresponding to the first state value in the segment from the timestamp corresponding to the last state value in the segment, and add a time sampling interval. The sampling interval is set at intervals (if equal intervals are used); if the sampling intervals are inconsistent, the actual time difference between adjacent valid state points within the segment needs to be accumulated; all consecutive "1" segments of the player are traversed to calculate multiple possible continuous occlusion duration values; the system usually records the maximum value among them, or selects one as the final continuous occlusion duration according to rules (such as the earliest continuous occlusion); if there is no segment with a state of "1" in the sequence, the continuous occlusion duration is zero; the continuous time period can be a pre-set analysis window, or it can be dynamically determined according to the game events; as a preferred embodiment, the continuous time period usually begins after the serve and ends before the offensive action (such as the spike) or before the ball crosses the net; in other embodiments, a sliding time window can also be used for real-time analysis, which is not limited in this application.

[0057] It should be noted that the continuous occlusion duration mentioned in this application refers to the physical time length corresponding to a continuous segment with a state of "occlusion" (such as "1") calculated from a continuous binary state sequence.

[0058] In step 105, if the duration of continuous obstruction exceeds a preset threshold, and the player is determined to have a positional error as defined by the volleyball rules based on the relative positional relationship, then a violation of the cover position rule is determined to have occurred.

[0059] In some embodiments, if the duration of continuous obstruction exceeds a preset threshold, and the player is determined to have committed a positional error as defined by the volleyball rules based on the relative positional relationship, the determination of a cover-up violation can be achieved through the following steps: The duration of continuous occlusion is compared with a preset threshold to determine whether the timeout condition is met. Based on the relative positional relationship, determine whether the non-offensive core player who caused the continuous blocking duration constitutes a positional error as defined by the volleyball rules; If both the timeout condition and the position error condition are met simultaneously, it is determined that the non-offensive core player has violated the cover positioning rule.

[0060] For specific implementation, refer to Figure 3 As shown in the figure, this is a flowchart illustrating the process of determining timeout conditions in some embodiments of this application. The determination of whether the timeout condition is met by comparing the continuous obstruction duration with a preset threshold can be achieved in the following ways: For example, obtaining a preset threshold, which is a time threshold value preset based on the penalty standards for covert positioning violations in volleyball matches, after statistical analysis of a large amount of historical match video data; the threshold can be adjusted according to the match level, the strictness of the referee, or specific match rules, with a larger value indicating a more lenient system penalty and a smaller value indicating a more strict system penalty; directly comparing the continuous obstruction duration of the team's front-row non-offensive core player with the preset threshold value: if the continuous obstruction duration is greater than the preset threshold, the timeout condition is determined to be met; if it is less than or equal to, the timeout condition is determined not to be met.

[0061] It should be noted that the timeout condition mentioned in this application refers to a Boolean logic judgment result obtained by comparing whether the duration of continuous occlusion is greater than a preset time threshold.

[0062] In specific implementation, based on the relative positional relationship, determining whether a non-offensive core player who caused the continuous obstruction duration constitutes a positional error as defined by volleyball rules can be achieved in the following way: First, obtain the relative positional relationship based on the non-offensive core player's identification; the relative positional relationship includes the quantified distance and orientation information of the player's feet and key court landmarks (such as the center line and the three-meter line) at the instant before the serve; then, according to the definition of "positional error" in the official competition rules of the International Volleyball Federation, convert the player's relative positional relationship data into a rule judgment; the judgment is mainly based on the rotation order: by comparing the positional relationship of the player's feet with the feet of adjacent players in the same row or column in the world coordinate system (e.g., a front-row player's feet must not be closer to the right sideline than the feet of the player to their right in the same row, and part of their feet must not cross the center line into the opponent's court, etc.), and automatically adjudicating based on the initial positioning matrix before the serve using programmed logical conditions; if the relative positional relationship data violates any relevant rule conditions, the player is determined to have committed a positional error; otherwise, it is not.

[0063] It should be noted that the positional error mentioned in this application refers to a violation of the rotation order or foot position determined according to the rules of volleyball based on the player's initial stance (i.e., the relative positional relationship) at the moment before the serve. Its physical significance is to provide a rule-based prerequisite for judging a cover position violation, that is, the player is already in a violation state in the initial stance.

[0064] In specific implementation, if both the timeout condition and the position error condition are met simultaneously, the determination that the non-offensive core player has committed a cover positioning violation can be achieved in the following way: For example, receive the determination results from the first two sub-steps: the first is the Boolean value (true or false) of the timeout condition, and the second is the Boolean value (true or false) of the position error condition; perform a logical AND operation: the output of the logical AND operation is "true" if and only if both input values ​​are "true"; if the output is "true", the system finally determines that the specific non-offensive core player has committed a "cover positioning violation" in the current game segment; subsequently, the system can generate a violation event record containing the player's identity, the time of the violation, the duration of continuous obstruction, and the specific type of violation, which can be used to trigger real-time alarms, generate penalty suggestions, or be presented in the post-game report; if the output of the logical AND operation is "false", it is determined that the violation did not occur; as a preferred embodiment, the above determination process can be executed in parallel for all monitored non-offensive core players to achieve synchronous monitoring of potential violations by multiple players on the field.

[0065] It should be noted that the cover positioning violation mentioned in this application refers to a specific game violation event judgment generated when the system simultaneously determines that a non-offensive core player meets the timeout condition and constitutes the positional error. Its physical meaning is to automatically identify the violation behavior of athletes using illegal initial positioning and continuously blocking the opponent's line of sight to cover the attack through technical means.

[0066] Furthermore, in another aspect of this application, in some embodiments, this application provides an intelligent detection system for abnormal behavior in sports events, including an intelligent detection unit for violations of positioning rules, as referenced. Figure 4 The figure is a schematic diagram of the structure of a station violation intelligent detection unit 200 according to some embodiments of this application. The station violation intelligent detection unit 200 includes: an acquisition module 201, a processing module 202, and an execution module 203, which are described below: The acquisition module 201 in this application is mainly used to acquire video images of the field at the moment before the serving player hits the ball, and to identify the relative positional relationship between the feet of the front row players of both sides and the field markings. Processing module 202, in this application, is mainly used to track the head position of the opposing setter based on the video image of the court from the moment of serving and hitting the ball, and then estimate the corresponding line of sight, and simultaneously track the real-time position of all attacking candidates in the front row of the team. In addition, the processing module 202 in this application is also used to determine the attacking candidate players in the extended area of ​​the line of sight as the observed attacking reference point based on the line of sight and the real-time position of all attacking candidate players, thereby constructing a dynamic spatial connection line from the head of the setter to the attacking reference point. In addition, the processing module 202 in this application is also used to monitor the movement of the non-offensive core players in the front row of the team based on the dynamic spatial connection, and then determine the continuous occlusion duration of the body envelope of the non-offensive core players in the front row of the team that continuously occludes the dynamic spatial connection through three-dimensional spatial analysis. The execution module 203 in this application is mainly used to determine that a cover position violation has occurred if the duration of continuous obstruction exceeds a preset threshold and the player is judged to have committed a positional error as defined by the volleyball rules based on the relative positional relationship.

[0067] In addition, this application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described intelligent detection method for volleyball positioning violations.

[0068] In some embodiments, reference Figure 5 This figure is an internal structural diagram of a computer device implementing an intelligent detection method for volleyball positioning violations according to some embodiments of this application. The intelligent detection method for volleyball positioning violations in the above embodiments can... Figure 5 The computer device shown is used to implement this, and the computer device 300 includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0069] The processor 301 may be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices used to control the execution of the intelligent detection method for volleyball positioning violations in this application.

[0070] The communication bus 302 is used to transmit information between the aforementioned components.

[0071] Memory 303 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 303 may exist independently and be connected to processor 301 via communication bus 302. Memory 303 may also be integrated with processor 301.

[0072] The memory 303 stores program code for executing the solution of this application, and its execution is controlled by the processor 301. The processor 301 executes the program code stored in the memory 303. The program code may include one or more software modules. In the above embodiment, the intelligent detection method for volleyball positioning violations can be implemented by the processor 301 and one or more software modules in the program code in the memory 303.

[0073] Communication interface 304 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0074] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core processor or a multi-core processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0075] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device may be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0076] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described intelligent detection method for volleyball positioning violations.

[0077] In summary, the intelligent detection system and method for abnormal behavior in sports events disclosed in this application acquires a video image of the court just before the serving player hits the ball, and identifies the relative positional relationship between the feet of both teams' front-row players and the court markings; starting from the moment the ball is hit, the system tracks the head position of the opposing setter based on the video image of the court to estimate the corresponding line of sight, and simultaneously tracks the real-time positions of all attacking candidates in the front row of the team; based on the line of sight and the real-time positions of all attacking candidates, the attacking candidates within the extended area of ​​the line of sight are identified as the observed players. The system observes the attack reference point and constructs a dynamic spatial line from the setter's head to the attack reference point. Based on this dynamic spatial line, it monitors the movement of the team's non-offensive core players in the front row and determines the duration of continuous obstruction of the dynamic spatial line by the body envelope of the team's non-offensive core players in the front row through three-dimensional spatial analysis. If the duration of continuous obstruction exceeds a preset threshold and the player is judged to have a positional error as defined by the volleyball rules based on the relative positional relationship, a covert positioning violation is determined to have occurred. This system can effectively identify concealed and combined violations.

[0078] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0079] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for intelligent detection of volleyball positioning violations, applied to an intelligent detection system for abnormal behavior in sports events, characterized in that, The method includes the following steps: Just before the serving player hits the ball, acquire video images of the court and identify the relative positions of the feet of the front-row players from both sides to the court markings; Starting from the moment the ball is hit, the head position of the opposing setter is tracked based on the video images of the court to estimate the corresponding line of sight, and the real-time position of all attacking candidates in the front row of the team is tracked simultaneously. Based on the line of sight and the real-time positions of all attacking candidates, the attacking candidates within the extended area of ​​the line of sight are identified as the observed attacking reference points, thereby constructing a dynamic spatial connection from the setter's head to the attacking reference points. Based on the dynamic spatial connection, the movement of the team's front row non-offensive core players is monitored, and then the duration of continuous occlusion of the team's front row non-offensive core players' body envelope by the dynamic spatial connection is determined through three-dimensional spatial analysis. If the duration of continuous obstruction exceeds a preset threshold, and the player is judged to have a positional error as defined by the volleyball rules based on the relative positional relationship, then a violation of the cover position rule is determined to have occurred.

2. The method as described in claim 1, characterized in that, Starting from the moment the ball is struck during the serve, the method of tracking the head position of the opposing setter based on the court video images and then estimating the corresponding line of sight specifically includes: Starting from the moment the ball is served, the video images of the court are continuously analyzed frame by frame to maintain cross-frame identity tracking of the opposing setter. In each video frame, the head region of the opposing setter is detected and tracked to locate facial key points for posture estimation. Based on the image coordinates of the facial key points and their corresponding 3D head model, the head rotation posture is calculated, and then the corresponding gaze direction is estimated.

3. The method as described in claim 1, characterized in that, Synchronously tracking the real-time positions of all potential attacking players in the front row of our team specifically includes: Based on the athlete's initial identity associated with the relative positional relationship, multi-target tracking is initiated for the team's front-row players in the field video image; In the video images of the arena, the identification of each player in the front row of their team is maintained by associating the data of the detection box with the trajectory. For each front-line player on our side who successfully maintains their identity, the real-time positions of all attacking candidates on our side's front line are tracked through the world coordinate system mapping relationship.

4. The method as described in claim 1, characterized in that, Based on the line of sight and the real-time positions of all attacking candidates, the attacking candidates within the extended area of ​​the line of sight are identified as the observed attacking reference points. This process then constructs a dynamic spatial connection from the setter's head to the attacking reference points, specifically including: Based on the direction of the line of sight and the position of the setter's head in the world coordinate system, a regular spatial region is constructed with the position of the setter's head as the vertex and extending along the direction of the line of sight. The set of attacking candidate players located within the rule space area is obtained by filtering based on the real-time positions of all attacking candidate players; Based on preset selection rules, a unique player is selected from the set of attacking candidate players as the attacking reference point to be observed; Based on the real-time position coordinates of the setter's head and the attack reference point, a dynamic spatial connection is constructed from the setter's head to the attack reference point.

5. The method as described in claim 1, characterized in that, Monitoring the movement of our team's front-row non-offensive core players based on the aforementioned dynamic spatial connection specifically includes: Based on the dynamic spatial connection, the non-attack core players that need to be monitored are extracted from all the attacking candidate players in the front row of our side. Continuously acquire the real-time location coordinates of each non-offensive core player to monitor their movement trajectory.

6. The method as described in claim 1, characterized in that, The duration of continuous occlusion of the dynamic spatial connection line by the body envelope of the team's front-row non-offensive core player, determined through three-dimensional spatial analysis, specifically includes: A three-dimensional geometric model of the body envelope of each non-offensive core player is constructed based on the real-time position and posture of each player. By connecting the three-dimensional geometric model with the dynamic space, a three-dimensional occlusion determination is performed to obtain a binary state of whether occlusion has occurred at different time points; The duration of continuous occlusion of the dynamic spatial connection line by the body envelope of the non-offensive core player in the front row is determined based on the binary state within a continuous time period.

7. The method as described in claim 1, characterized in that, If the duration of continuous obstruction exceeds a preset threshold, and the player is determined to have committed a positional error as defined by the volleyball rules based on the relative positional relationship, then a violation of the cover positioning rule is determined to have occurred. Specifically, this includes: The duration of continuous occlusion is compared with a preset threshold to determine whether the timeout condition is met. Based on the relative positional relationship, determine whether the non-offensive core player who caused the continuous blocking duration constitutes a positional error as defined by the volleyball rules; If both the timeout condition and the position error condition are met simultaneously, it is determined that the non-offensive core player has violated the cover positioning rule.

8. An intelligent detection system for abnormal behavior in sports events, comprising an intelligent detection unit for violations of standing position rules, characterized in that, The intelligent detection unit for station violations includes: The acquisition module is used to acquire video images of the court just before the serving player hits the ball, and to identify the relative positional relationship between the feet of the front row players of both sides and the court markings. The processing module is used to track the head position of the opposing setter based on the video image of the court from the moment of the serve and hit the ball, and then estimate the corresponding line of sight, and simultaneously track the real-time position of all attacking candidates in the front row of the team. The processing module is also used to determine the attacking candidate players within the extended area of ​​the line of sight as the observed attacking reference points based on the line of sight and the real-time positions of all attacking candidate players, thereby constructing a dynamic spatial connection line from the setter's head to the attacking reference points. The processing module is also used to monitor the movement of the team's front row non-offensive core players based on the dynamic spatial connection, and then determine the duration of continuous occlusion of the dynamic spatial connection by the body envelope of the team's front row non-offensive core players through three-dimensional spatial analysis. The execution module is used to determine that a cover position violation has occurred if the duration of continuous obstruction exceeds a preset threshold and the player's position is determined to be incorrect according to the relative positional relationship as defined by the volleyball rules.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent detection method for volleyball positioning violations as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent detection method for volleyball positioning violations as described in any one of claims 1 to 7.

Images