Volleyball attack formation recognition method and system based on machine vision and GMM clustering algorithm

Abstract

The invention discloses a volleyball attack formation recognition method and system based on machine vision and a GMM clustering algorithm, and the method comprises the steps: obtaining a real-time video stream of a playing field, and recognizing a key frame through the analysis of a volleyball motion track and a turning point; performing target detection on the athlete of the key frame by using a YOLOv8 target detection algorithm, outputting a real-time position coordinate of the athlete, and performing frame compensation operation on the key frame to obtain continuous motion information; analyzing volleyball motion trend according to the continuous motion information to obtain volleyball motion parameters; clustering the real-time position coordinates of the athletes by using a Gaussian mixture model, dividing tactical grid regions and obtaining distribution characteristics of the athletes; the current athlete formation is judged according to the athlete distribution characteristics and the ball motion parameters, and an athlete formation optimization suggestion is output according to historical competition data. According to the method, the attack formation is judged and optimized by means of an automatic and objective way, and the limitation of an existing tactical analysis method is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a volleyball attack formation recognition method and system based on machine vision and GMM clustering algorithm. Background Art

[0002] In current volleyball matches, analyzing offensive formations is crucial for tactical planning and match decision-making. However, current tactical analysis often relies on manual judgment by coaches and tactical analysts based on their own experience. This presents several challenges. First, it is highly subjective. Manual analysis is constrained by personal experience and perspective, easily influenced by subjective factors, and lacks data support. Second, it suffers from poor real-time performance. In the high-speed competitive environment of a match, manual analysis struggles to accurately and promptly identify offensive formations and make effective tactical adjustments. Finally, data utilization is insufficient. Although modern games have introduced video analysis and sensor data collection technologies, existing methods are still inadequate for automated identification of offensive formations, lacking precise pattern recognition and quantitative analysis.

[0003] In view of the above problems, there is an urgent need for a volleyball offensive formation recognition method based on computer vision and data analysis, which can automatically and in real time analyze the offensive tactics of the game, accurately judge the offensive formation, and provide data support for tactical decision-making, thereby improving offensive efficiency and tactical execution. Summary of the Invention

[0004] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the existing technology and provide a volleyball offensive formation recognition method and system based on machine vision and GMM clustering algorithm. By integrating video data with motion trajectory information and applying machine learning and pattern recognition technology, real-time judgment of the offensive formation can be achieved. The system can be applied to various scenarios such as game tactical analysis, post-match review, and auxiliary coaching guidance.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm, comprising the following steps:

[0007] Obtain real-time video streams from the playing field and identify key frames by analyzing the volleyball's trajectory and turning points;

[0008] Use the YOLOv8 target detection algorithm to detect the athletes in the key frames, output the athletes' real-time position coordinates, and perform frame interpolation on the key frames to obtain continuous motion information. Analyze the volleyball movement trend based on the continuous motion information and obtain the ball movement parameters.

[0009] The Gaussian mixture model is used to cluster the real-time position coordinates of the players, divide the tactical grid area and obtain the player distribution characteristics;

[0010] Determine the current player formation based on player distribution characteristics and ball motion parameters, and output optimized player formation suggestions based on historical game data.

[0011] As a preferred technical solution, the method of identifying key frames by analyzing the volleyball's motion trajectory and turning points includes:

[0012] By analyzing the trajectory of the volleyball, the team identified key offensive moments, recording the coordinates and time of the ball frame by frame. These key moments included the passing moment, the take-off moment, and the spiking moment.

[0013] The analysis of the volleyball trajectory comprises the following steps:

[0014] Calculate the sphere movement speed V of consecutive frames as follows:

[0015]

[0016] Among them, (x1, y1) and (x2, y2) are the positions of the volleyball in two consecutive frames, and t2-t1 is the time difference, which is used to detect speed mutations and capture the starting point of the action;

[0017] The cosine of the angle θ between the two vectors is as follows:

[0018]

[0019] v1=[x i -x i+1 ,y i -y i+1 ],v2=[x i+1 -x i+2 ,y i+1 -y i+2 ];

[0020] Calculate the distance as follows:

[0021]

[0022] The key frames are filtered by combining speed and distance. When the volleyball speed does not exceed the set threshold and the distance is not less than the set threshold, the current two consecutive frames are marked as key frames.

[0023] As a preferred technical solution, the YOLOv8 target detection algorithm is used to detect the target of the athletes in the key frames, including

[0024] Extract the player's position from the video frame, calculate the center coordinates, and obtain the player's coordinates as follows:

[0025]

[0026] Where (x, y) is the coordinate of the upper left corner of the detection box, and h is the vertical size of the athlete in the image;

[0027] By adding frames that are continuous with the key frames before and after the key frames, continuous motion information is obtained, and the ball's speed and acceleration are calculated based on the continuous motion information. The continuous motion information includes the ball's speed, acceleration, and trajectory curvature. The offensive rhythm is analyzed by the ball's speed, the movement trend is analyzed by the acceleration, and the changing characteristics of the ball's movement path are analyzed by the trajectory curvature.

[0028] The distance constraint is combined with filtering the abnormal values ​​of the ball motion parameters, and the Kalman filter is used to reduce the noise of the key frames. The player coordinates are normalized based on the set resolution to obtain the distance constraint threshold that is dynamically adjusted over time.

[0029] As a preferred technical solution, the method of clustering the real-time position coordinates of the athletes using a Gaussian mixture model, dividing the tactical grid area and obtaining the distribution characteristics of the athletes includes:

[0030] The real-time position coordinates of all athletes in the current key frame are mapped to the court plane through perspective transformation. The transformation formula is:

[0031] [x′,y′,1]=M·[x·1920, (y+h / 2)·1080-821,1];

[0032] Where x' is the horizontal coordinate of the athlete's mapped coordinates, y' is the vertical coordinate of the athlete's mapped coordinates, and M is the coefficient of variation;

[0033] The court plane is set as a 3*3 grid and the court is allocated with tactical grid areas. The Gaussian mixture model is used to cluster the mapping coordinates of the players and mark the distribution characteristics of the players.

[0034] As a preferred technical solution, the tactical grid area allocation of the court is as follows:

[0035] As a preferred technical solution, the athlete distribution characteristics include front row, middle row and back row.

[0036] As a preferred technical solution, the method of determining the current player formation based on player distribution characteristics and ball motion parameters includes:

[0037] Calculate the difference in attack time between athletes and judge the speed of attack;

[0038] By detecting the offensive behavior of athletes, we can judge the offensive intensity;

[0039] According to the speed and intensity of the offensive rhythm, combined with the ball movement parameters, the player distribution is matched with the preset standard formation to analyze and judge the current player formation.

[0040] As a preferred technical solution, the method of detecting the athlete's offensive behavior includes: judging whether the athlete has taken off based on the vertical displacement and horizontal distance, as shown in the following formula:

[0041] Vertical displacement:

[0042] Horizontal displacement: Distance x <β;

[0043] Among them, |y i+1 -y i | is the vertical displacement of the athlete in consecutive frames, α is the jump threshold, and β is the horizontal distance threshold;

[0044] The athlete's arm swing action is compared and analyzed with the pre-defined standard spiking action pattern. When the similarity exceeds the set threshold, it is determined that the athlete has a spiking action.

[0045] As a preferred technical solution, the outputting of optimized player formation suggestions based on historical game data includes:

[0046] Based on the current player formation, query historical match data and corresponding historical formations under the same circumstances, compare and score the historical formations with the current formation, and calculate the position score; the historical match data includes the attack success rate and the corresponding historical formations of successful attacks;

[0047] Calculate the current formation score as follows:

[0048] Score=ω1×attack success rate+ω2×player position score;

[0049] Among them, attack success rate is the attack success rate, player position score is the position score, ω1 and ω2 are weight coefficients;

[0050] Based on the current formation score, the offensive mode of both teams' formations is identified and whether tactical adjustments are needed. If the score is lower than the set threshold, optimization suggestions are output, including suggestions for adjusting athlete positions, optimizing offensive routes, adjusting tactical combinations, and targeted defensive breakthroughs.

[0051] In a second aspect, the present invention further provides a volleyball offensive formation recognition system based on machine vision and GMM clustering algorithm, which is applied to the volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm, and includes a data processing module, a target detection module, a position division module, and an optimization decision module;

[0052] A data processing module is used to acquire a real-time video stream of the playing field using a high-frame-rate camera or sensor, and to identify key frames by analyzing the volleyball's trajectory and turning points;

[0053] The target detection module uses the YOLOv8 target detection algorithm to detect the players in the key frames, output the players' real-time position coordinates, and perform frame interpolation operations on the key frames to obtain continuous motion information. Based on the continuous motion information, the module analyzes the volleyball movement trend and obtains the ball movement parameters.

[0054] Position segmentation module, used to cluster the real-time position coordinates of athletes using Gaussian mixture model, divide the tactical grid area and obtain the distribution characteristics of athletes;

[0055] The optimization decision module is used to determine the current player formation based on the player distribution characteristics and ball movement parameters, and output optimized player formation suggestions based on historical game data.

[0056] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0057] Compared to traditional tactical analysis methods, this invention offers the following advantages: First, it is highly real-time, capturing key frames in real time during the offensive phase of a match and quickly identifying offensive formations, providing timely data support for tactical adjustments. Second, it is highly accurate, integrating computer vision and machine learning technologies to improve the accuracy of offensive formation identification and reduce human misjudgment. Third, it is objective and automated, reducing reliance on human experience, providing data-driven tactical decision support, and improving the scientific nature and stability of analysis. Fourth, it provides tactical optimization support, leveraging historical data analysis to provide offensive formation optimization suggestions and enhance the effectiveness of tactical execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0059] Figure 1 This is a flow chart of a volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm according to an embodiment of the present invention;

[0060] Figure 2 This is a schematic diagram of the key frame detection process according to an embodiment of the present invention;

[0061] Figure 3 This is a schematic diagram of the process of collecting real-time position data and extracting indicators of athletes according to an embodiment of the present invention;

[0062] Figure 4 Schematic diagram of the clustering and station analysis process based on GMM in an embodiment of the present invention;

[0063] Figure 5 The figure is a structural diagram of a volleyball offensive formation recognition system based on machine vision and GMM clustering algorithm according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0065] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments.

[0066] The time and sequence expressions such as "current", "next" or "next" described in this embodiment are expressed based on a certain key frame obtained as a reference system. The key frame contains information such as the players, the ball, the court and their relative positions. By linking the relationship between the key frame and other continuous video frames, it can be used as the basis for identifying the position of the volleyball offensive formation in this embodiment or specifying a certain object.

[0067] See also Figure 1 This embodiment provides a volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm, comprising the following steps:

[0068] S1. Use a high-frame-rate camera or sensor to obtain a real-time video stream of the competition venue, and identify key frames by analyzing the volleyball's trajectory and turning points.

[0069] Step S1 uses a high-frame-rate camera to capture video of the volleyball game, continuously tracks the ball's trajectory with the help of image processing algorithms, analyzes the turning points of the trajectory, extracts key frames in the attack process, such as the start time of the attack and the attack contact point, and uses these key moments as opportunities for subsequent data analysis.

[0070] In the process of tracking the ball's trajectory, this embodiment identifies the key moments of attack by analyzing the volleyball's trajectory and records the coordinates and time of the ball frame by frame. The key moments can be divided into the passing moment, the take-off moment, and the spiking moment.

[0071] It's important to clarify that the passing moment refers to the moment or frame when the player's arm or hand contacts the ball during a pass; the take-off moment refers to the moment or frame when the player takes off; and the spike moment refers to the moment or frame when the player's arm swing closely resembles or aligns with a standard spike, making contact with the ball. These three key moments represent typical turning points in volleyball techniques and are crucial for accurately identifying offensive formations.

[0072] How to analyze the trajectory of volleyball, such as Figure 2 As shown, the following steps are included:

[0073] S11. Calculate the ball movement speed V of consecutive frames as follows:

[0074]

[0075] Among them, (x1, y1) and (x2, y2) are the positions of the volleyball in two consecutive frames, and t2-t1 is the time difference, which is used to detect speed mutations and capture the starting point of the action;

[0076] S12, through the vector angle cosine θ, as follows:

[0077]

[0078] v1=[x i -x i+1 ,y i -y i+1 ],v2=[x i+1 -x i+2 ,y i+1 -y i+2 ];

[0079] S13. Calculate the distance as follows:

[0080]

[0081] Combine speed and angle to filter key frames. When the volleyball speed does not exceed the set threshold and the distance is not less than the set threshold, the two consecutive frames are marked as a group of key frames as follows: V≤V limit , Distance≥D limit .

[0082] In actual implementation, the first 15 frames speed threshold V limit =0.65, distance threshold D limit =4.5, the subsequent frames are adjusted to V limit =0.4, D limit =8, dynamically adapting to the trajectory characteristics at different stages of the game to ensure accuracy.

[0083] It should be noted that key frames are marked and processed in units of two consecutive frames rather than a single independent frame, which helps to capture the transition characteristics of the ball's motion changes.

[0084] S2. Use the YOLOv8 target detection algorithm to detect the athletes in the key frames, output the athletes' real-time position coordinates, perform frame interpolation operations on the key frames, and obtain continuous motion information; analyze the volleyball movement trend based on the continuous motion information and obtain the ball movement parameters.

[0085] In this embodiment, step S2 firstly, this embodiment needs to perform target detection on the athletes in the key frames using the YOLOv8 target detection algorithm, including the following steps:

[0086] S21. Extract the player's position from the video frame, calculate the center coordinates, and obtain the player's coordinates as follows:

[0087]

[0088] Where (x, y) is the coordinate of the upper left corner of the detection box, and h is the vertical size of the athlete in the image.

[0089] S22. Continuous motion information is obtained by adding frames that are continuous with the key frame before and after the key frame. The continuous motion information specifically includes characteristic parameters of the ball's motion, such as the ball's speed, acceleration, and trajectory curvature. The ball's speed and acceleration are calculated based on this continuous motion information. The offensive rhythm is analyzed using the ball's speed, the motion change trend is analyzed using the acceleration, and the turning characteristics of the ball's motion path are analyzed using the trajectory curvature, as shown in the following formula:

[0090] speed acceleration

[0091] It's worth explaining that at this specific moment in the key frame, target detection technology or a sensor system is used to collect real-time position data for each offensive player, specifically the x and y coordinates. This continuous motion information includes data such as the ball's speed, acceleration, and trajectory curvature. All of this data undergoes filtering, noise reduction, and normalization to ensure accuracy and consistency, as described in step S23.

[0092] S23. Filter outliers of the ball motion parameters in combination with the distance constraint, use Kalman filtering to reduce noise on key frames, and normalize the player coordinates based on a set resolution to obtain the distance constraint threshold that is dynamically adjusted over time.

[0093] The distance constraint can be expressed as Distance<max(min(0.15·Δt, 0.6), 0.15). The distance constraint threshold is dynamically adjusted over time, with a maximum value of 0.6 and a minimum value of 0.15 to ensure detection stability.

[0094] In addition, for unified analysis, the data were preprocessed using Kalman filtering for noise reduction, and the coordinates were normalized to the range of 0 to 1 based on the 1920 × 1080 resolution.

[0095] It can be explained that the Kalman filter is an efficient recursive algorithm for estimating the state of a dynamic system from noisy measurement data. It combines two steps: prediction and measurement update: (1) prediction phase: predicting the state and error covariance at the next moment based on the system model; (2) update phase: combining the actual measurement data, revising the predicted value to obtain a more accurate state estimate.

[0096] S3. Use the Gaussian mixture model to cluster the real-time position coordinates of the athletes, divide the tactical grid area and obtain the athlete distribution characteristics.

[0097] like Figure 3 As shown, in this embodiment, the real-time position coordinates of all athletes in the current key frame are mapped to the court plane through perspective transformation. The transformation formula is:

[0098] [x′,y′,1]=M·[x·1920,(y+h / 2)·1080-821,1];

[0099] Where x' is the horizontal coordinate of the player's mapped coordinates, y' is the vertical coordinate of the player's mapped coordinates, and M is the coefficient of variation. M is calculated from the source points [[0, 0], [1, 0], [0, 1], [1, 1]] and the target points [[0, 0], [1800, 0], [0, 900], [1800, 900]] to convert the video coordinates into the court coordinate system (1800 pixels wide × 900 pixels high).

[0100] The court plane is set as a 3*3 grid and the court is allocated with tactical grid areas. The Gaussian mixture model is used to cluster the mapping coordinates of the players and mark the distribution characteristics of the players.

[0101] Then, if Figure 4 As shown, Gaussian mixture model GMM clustering is used, and the probability density The front row, middle row and back row distribution are identified by 9 cluster centers. To simplify the real-time calculation, the court is divided into a 3×3 grid. The area allocation formula is: Quickly locate the player's position and analyze the position characteristics based on clustering results.

[0102] In the specific implementation process, the cluster number K in this embodiment is 9, the grid width is 600 pixels, and the height is 300 pixels to ensure uniform partitioning.

[0103] In the above step S3, the pre-processed position data is input into the Gaussian mixture model (GMM) to carry out cluster analysis. With the help of the cluster centers obtained by the analysis, the distribution of players in the front row, middle row and back row is clarified, and compared with the pre-set standard formation to preliminarily determine the current formation type.

[0104] S4. Determine the current player formation based on player distribution characteristics and ball motion parameters, and output optimized player formation suggestions based on historical game data.

[0105] After the above-mentioned S2-S3 calculation and analysis of the player distribution characteristics, this embodiment can detect the speed and intensity of the offensive rhythm by calculating the time difference and displacement, and further analyze the current player formation in combination with the ball movement parameters. Specifically, it includes the following steps:

[0106] S41. Calculate the time difference between athletes’ attacks and judge the speed of the attack.

[0107] Here, it can be specifically understood that if the time difference or the difference between the continuous frame and the key frame is small, it is a fast attack, otherwise it can be understood as defense or stillness. These other situations are not the subject of research. In a specific application, it can be set that the time difference is less than 60 frames as a fast attack.

[0108] S42. Determine the offensive intensity by detecting the athlete's offensive behavior.

[0109] In step S42, according to the general understanding of volleyball, offensive behaviors can be divided into passing, jumping, swinging, spiking, blocking, etc. This embodiment mainly takes jumping, swinging, and spiking as research objects, and carries out the following identification steps:

[0110] S421. Determine whether the athlete has taken off based on vertical displacement and horizontal distance, as follows:

[0111] Vertical displacement:

[0112] Horizontal displacement: Distance x <β;

[0113] Among them, |y i+1 -y i | is the vertical displacement of the athlete in consecutive frames, α is the jump threshold, and β is the horizontal distance threshold. You can set the jump threshold to 0.02 and the horizontal distance threshold to 0.055.

[0114] It should be noted that the jumping action does not necessarily lead to a spiking action. The athlete may spike the ball after jumping, or may perform a jump serve or jump pass. Therefore, it is necessary to further identify whether the arm swing action is a spiking action, see step S422 for details.

[0115] S422: Compare and analyze the athlete's arm swing with a predefined standard spike pattern. If the similarity exceeds a set threshold, the athlete is judged to have performed a spike. This determination is independent of the jump determination, and the combination of the two can distinguish different technical movements such as spikes and jump passes.

[0116] S43. Based on the speed and intensity of the offensive rhythm and in combination with the ball movement parameters, the player distribution is matched with the preset standard formation to analyze and determine the current player formation.

[0117] Finally, based on the current player formation, historical match data and corresponding historical formations under the same circumstances are retrieved. The historical formations are compared and scored against the current formation to calculate a positional score. The historical match data includes the attack success rate and the corresponding historical formations of successful attacks. The attack success rate refers to the ratio of completed attacks and scores in the same or similar formations in the historical match data. The positional score is calculated by comparing the current player's position with the position of the historical successful formations by calculating the spatial distance and posture similarity, reflecting the degree of match between the current formation and the historical successful formations.

[0118] Calculate the current formation score as follows:

[0119] Score=0.6×attack success rate+0.4×player position score;

[0120] The attack success rate is the attack success rate, and the player position score is the position score. In practice, they are set to 0.6 and 0.4, respectively, to balance the impact of historical success rate and current position matching.

[0121] Based on the current formation score, the attacking patterns of both teams are identified and whether tactical adjustments are needed. If the score is lower than the set threshold, optimization suggestions are output.

[0122] In this embodiment, the system will output the following types of optimization suggestions based on the comparison results of the current formation score and the set threshold:

[0123] (1) Athlete Position Adjustment Suggestions: Based on historically successful formations, provide specific positioning adjustments for athletes in specific areas of the court;

[0124] (2) Offensive route optimization suggestions: Analyze the offensive routes with high success rates in historical games and provide the optimal offensive path selection based on the current situation on the field;

[0125] (3) Tactical combination adjustment suggestions: Based on the formation patterns in historical game data that are similar to the current situation but have a higher success rate, provide optimization solutions for the overall tactical combination;

[0126] (4) Targeted defensive breakthrough suggestions: Based on the characteristics of the opponent's current defensive formation and combined with successful breakthrough points in similar situations in historical games, provide targeted offensive tactical suggestions.

[0127] These suggestions are based on successful formation patterns statistically analyzed from historical match data and are dynamically generated through real-time analysis of the current on-field situation. It should be noted that for the aforementioned method embodiments, for ease of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the present invention is not limited to the order of the actions described, as certain steps can be performed in other orders or simultaneously according to the present invention.

[0128] Based on the same concept as the volleyball offensive formation recognition method based on machine vision and a GMM clustering algorithm described in the aforementioned embodiment, the present invention also provides a volleyball offensive formation recognition system based on machine vision and a GMM clustering algorithm, which can be used to execute the aforementioned volleyball offensive formation recognition method based on machine vision and a GMM clustering algorithm. For ease of explanation, the structural diagram of the embodiment of the volleyball offensive formation recognition system based on machine vision and a GMM clustering algorithm only shows the parts relevant to the embodiment of the present invention. Those skilled in the art will understand that the illustrated structure does not constitute a limitation of the device, and the device may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0129] See also Figure 5 In another embodiment of the present application, a volleyball offensive formation recognition system 10 based on machine vision and GMM clustering algorithm is provided, the system comprising a data processing module 11, a target detection module 12, a position division module 13 and an optimization decision module 14;

[0130] The data processing module 11 is used to obtain a real-time video stream of the playing field using a high-frame-rate camera or sensor, and identify key frames by analyzing the volleyball's trajectory and turning points;

[0131] The target detection module 12 is used to detect the players in the key frames using the YOLOv8 target detection algorithm, output the real-time position coordinates of the players, perform frame interpolation operations on the key frames, and obtain continuous motion information; analyze the volleyball movement trend based on the continuous motion information to obtain the ball movement parameters;

[0132] A position division module 13 is used to cluster the real-time position coordinates of the athletes using a Gaussian mixture model, divide the tactical grid area and obtain the distribution characteristics of the athletes;

[0133] Determine the current player formation based on player distribution characteristics and ball motion parameters, and output optimized player formation suggestions based on historical game data.

[0134] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0135] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm, characterized in that: The steps include: Obtain real-time video streams from the playing field and identify key frames by analyzing the volleyball's trajectory and turning points; Use the YOLOv8 target detection algorithm to detect the athletes in the key frames, output the athletes' real-time position coordinates, and perform frame interpolation on the key frames to obtain continuous motion information; Analyze volleyball movement trends based on continuous movement information to obtain ball movement parameters; The Gaussian mixture model is used to cluster the real-time position coordinates of the players, divide the tactical grid area and obtain the player distribution characteristics; Determine the current player formation based on player distribution characteristics and ball motion parameters, and output optimized player formation suggestions based on historical game data.

2. The volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm according to claim 1 is characterized in that: The method of identifying key frames by analyzing the volleyball trajectory and turning points includes: By analyzing the trajectory of the volleyball, the team identified key offensive moments, recording the coordinates and time of the ball frame by frame. These key moments included the passing moment, the take-off moment, and the spiking moment. The analysis of the volleyball trajectory comprises the following steps: Calculate the sphere movement speed V of consecutive frames as follows: Among them, (x1, y1) and (x2, y2) are the positions of the volleyball in two consecutive frames, and t2-t1 is the time difference, which is used to detect speed mutations and capture the starting point of the action; The cosine of the angle θ between the two vectors is as follows: v1=[x i -x i+1 ,y i -y i+1 ],v2=[x i+1 -x i+2 ,y i+1 -y i+2 ]; Calculate the distance as follows: The key frames are filtered by combining speed and distance. When the volleyball speed does not exceed the set threshold and the distance is not less than the set threshold, the current two consecutive frames are marked as key frames.

3. The volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm according to claim 1 is characterized in that: The YOLOv8 target detection algorithm is used to detect the target of the key frame athletes, including Extract the player's position from the video frame, calculate the center coordinates, and obtain the player's coordinates as follows: Where (x, y) is the coordinate of the upper left corner of the detection box, and h is the vertical size of the athlete in the image; By adding frames that are continuous with the key frames before and after the key frames, continuous motion information is obtained, and the ball's speed and acceleration are calculated based on the continuous motion information. The continuous motion information includes the ball's speed, acceleration, and trajectory curvature. The offensive rhythm is analyzed by the ball's speed, the movement trend is analyzed by the acceleration, and the changing characteristics of the ball's movement path are analyzed by the trajectory curvature. The distance constraint is combined with filtering the abnormal values ​​of the ball motion parameters, and the Kalman filter is used to reduce the noise of the key frames. The player coordinates are normalized based on the set resolution to obtain the distance constraint threshold that is dynamically adjusted over time.

4. The volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm according to claim 1 is characterized in that: The method of clustering the real-time position coordinates of the players using the Gaussian mixture model, dividing the tactical grid area and obtaining the distribution characteristics of the players includes: The real-time position coordinates of all athletes in the current key frame are mapped to the court plane through perspective transformation. The transformation formula is: [x′,y′,1]=M·[x·1920, (y+h / 2)·1080-821,1]; Where x' is the horizontal coordinate of the athlete's mapped coordinates, y' is the vertical coordinate of the athlete's mapped coordinates, and M is the coefficient of variation; The court plane is set as a 3*3 grid and the court is allocated with tactical grid areas. The Gaussian mixture model is used to cluster the mapping coordinates of the players and mark the distribution characteristics of the players.

5. The volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm according to claim 4 is characterized in that: The tactical grid area allocation for the court is as follows:

6. The volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm according to claim 4 is characterized in that: The athlete distribution characteristics include the front row, middle row and back row.

7. The volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm according to claim 1 is characterized in that: The determining of the current player formation according to the player distribution characteristics and the ball motion parameters includes: Calculate the difference in attack time between athletes and judge the speed of attack; By detecting the offensive behavior of athletes, we can judge the offensive intensity; According to the speed and intensity of the offensive rhythm, combined with the ball movement parameters, the player distribution is matched with the preset standard formation to analyze and judge the current player formation.

8. The volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm according to claim 1 is characterized in that: The detecting of the athlete's offensive behavior includes: judging whether the athlete has taken off based on the vertical displacement and horizontal distance, as shown in the following formula: Vertical displacement: Horizontal displacement: Distance x <β; Among them, |y i+1 -y i | is the vertical displacement of the athlete in consecutive frames, α is the jump threshold, and β is the horizontal distance threshold; The athlete's arm swing action is compared and analyzed with the pre-defined standard spiking action pattern. When the similarity exceeds the set threshold, it is determined that the athlete has a spiking action.

9. The volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm according to claim 1 is characterized in that: The output of optimized player formation suggestions based on historical game data includes: Based on the current player formation, query historical match data and corresponding historical formations under the same circumstances, compare and score the historical formations with the current formation, and calculate the position score; the historical match data includes the attack success rate and the corresponding historical formations of successful attacks; Calculate the current formation score as follows: Score=ω1×attack success rate+ω2×player position score; Among them, attack success rate is the attack success rate, player position score is the position score, ω1 and ω2 are weight coefficients; Based on the current formation score, the offensive mode of both teams' formations is identified and whether tactical adjustments are needed. If the score is lower than the set threshold, optimization suggestions are output, including suggestions for adjusting athlete positions, optimizing offensive routes, adjusting tactical combinations, and targeted defensive breakthroughs.

10. A volleyball offensive formation recognition system based on machine vision and GMM clustering algorithm, characterized by: A volleyball offensive formation recognition method based on machine vision and GMM clustering algorithm applied to any one of claims 1-9, comprising a data processing module, a target detection module, a position division module, and an optimization decision module; A data processing module is used to acquire a real-time video stream of the playing field using a high-frame-rate camera or sensor, and to identify key frames by analyzing the volleyball's trajectory and turning points; The target detection module is used to detect the athletes in the key frames using the YOLOv8 target detection algorithm, output the real-time position coordinates of the athletes, and perform frame interpolation operations on the key frames to obtain continuous motion information; Analyze volleyball movement trends based on continuous movement information to obtain ball movement parameters; Position segmentation module, used to cluster the real-time position coordinates of athletes using Gaussian mixture model, divide the tactical grid area and obtain the distribution characteristics of athletes; The optimization decision module is used to determine the current player formation based on the player distribution characteristics and ball movement parameters, and output optimized player formation suggestions based on historical game data.

Images