Sports event risk prediction method and system based on image vision

Abstract

The invention provides a sports event risk prediction method and system based on image vision, and relates to the technical field of sports event risk prediction based on image vision. The method comprises the following steps: acquiring a historical mating video of a match opponent of an athlete, a pre-match training video of the athlete and shoulder muscle information after the pre-match training of the athlete is finished; determining shoulder multi-time stretching information of each hand-hand simulation video by using a stretching information processing model based on the information of the plurality of hand-hand simulation videos of the game; determining a shoulder strain probability distribution diagram of each hand-handed simulation video based on the shoulder muscle soreness degree distribution diagram and the shoulder multi-time stretching information of each hand-handed simulation video; and determining the shoulder strain risk value of the competition for the shoulder strain probability distribution diagram of each hand-handing simulation video based on the graph neural network. The method can accurately pre-judge the shoulder strain risk of the competition in combination with the hand-handing action characteristics and the state of the athlete.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image vision-based sports event risk prediction, in particular to an image vision-based sports event risk prediction method and system. BACKGROUND

[0002] In badminton, tennis and other two-person confrontation sports events, the risk of shoulder muscle injury of athletes is a key problem affecting the performance of the competition. Athletes' shoulders need to frequently complete high-intensity and high-frequency actions such as serving, hitting, and defending. Long-term training and competition accumulation of muscle fatigue and action load can easily lead to sports injuries such as shoulder strain, which not only may interrupt the current event, but also may cause chronic injuries and affect the long-term competitive state of athletes. In the current sports event preparation stage, the evaluation of the risk of shoulder injury of athletes mainly relies on the experience analysis and judgment of coaches or the monitoring of basic physiological indicators, which has obvious limitations. Artificial analysis relies too much on experience and needs to invest a lot of time to review historical competition and training videos and organize data, and also has strong subjectivity. The evaluation result is easily affected by individual cognition and experience accumulation difference, and it is difficult to form unified and objective judgment standards. Traditional methods can only consider the training state of athletes or the tactical style of opponents, and cannot fully combine the two, which cannot accurately identify high-risk action confrontation scenes in the competition, and also cannot predict the specific load of the scenes on the shoulders of athletes. The existing scheme cannot combine the pre-competition muscle state of athletes and the action characteristics of opponents for systematic analysis, which leads to insufficient accuracy of strain risk assessment, and the risk warning lacks scientificity and forward-looking, and finally cannot provide accurate competition decision reference or targeted protection suggestions for athletes.

[0003] Therefore, how to accurately predict the risk of shoulder strain in the competition by combining the action characteristics of opponents and the state of athletes is a problem to be solved. SUMMARY

[0004] The technical problem solved by the present application is how to accurately predict the risk of shoulder strain in the competition by combining the action characteristics of opponents and the state of athletes.

[0005] According to a first aspect, the present application provides a sports event risk prediction method based on image vision, comprising: acquiring historical match video of an athlete competing against an opponent, pre-match training video of the athlete, and shoulder muscle information after pre-match training of the athlete; generating a plurality of match simulation video information of the current match based on the historical match video of the athlete competing against the opponent, the pre-match training video of the athlete, and the shoulder muscle information after pre-match training of the athlete; determining a shoulder muscle soreness distribution map based on the shoulder muscle information after pre-match training of the athlete; determining shoulder multiple stretching information of each match simulation video using a stretching information processing model based on the plurality of match simulation video information of the current match; determining a shoulder strain probability distribution map of each match simulation video based on the shoulder muscle soreness distribution map and the shoulder multiple stretching information of each match simulation video; determining a shoulder strain risk value of the current match based on a graph neural network for the shoulder strain probability distribution map of each match simulation video; and if the shoulder strain risk value of the current match is greater than a preset threshold, sending a risk warning signal to a control terminal.

[0006] In a possible implementation, the determining, by the graph neural network, of the shoulder strain risk value of the current match based on the shoulder strain probability distribution map of each match simulation video comprises: constructing a strain analysis graph, the strain analysis graph comprising a plurality of match simulation nodes and a plurality of edges between the plurality of match simulation nodes, a node feature of each match simulation node comprising each match simulation video information, the shoulder strain probability distribution map of each match simulation video, the shoulder muscle information after pre-match training of the athlete, and the shoulder muscle soreness distribution map, and an edge between simulation nodes being a credibility difference value of a match simulation video; and processing, by the graph neural network, the strain analysis graph to obtain the shoulder strain risk value of the current match.

[0007] In a possible implementation, the determining, based on the shoulder muscle soreness distribution map and the shoulder multiple stretching information of each match simulation video, of the shoulder strain probability distribution map of each match simulation video comprises: clustering, based on the shoulder multiple stretching information of each match simulation video, to obtain K clusters; determining a plurality of high-risk stretching information based on the K clusters; and determining, based on the plurality of high-risk stretching information and the shoulder muscle soreness distribution map, the shoulder strain probability distribution map of each match simulation video.

[0008] In a possible implementation, each match simulation video information of the current match comprises a match simulation video and a credibility of the match simulation video.

[0009] According to a second aspect, the present application provides a sports event risk prediction system based on image vision, comprising: an acquisition module configured to acquire historical match video of an athlete competing against an opponent, pre-match training video of the athlete, and shoulder muscle information of the athlete after pre-match training; a simulation module configured to generate a plurality of match simulation video information of the current match based on the historical match video of the athlete competing against the opponent, the pre-match training video of the athlete, and the shoulder muscle information of the athlete after pre-match training; an acid swelling degree distribution map determination module configured to determine a shoulder muscle acid swelling degree distribution map based on the shoulder muscle information of the athlete after pre-match training; a stretching information determination module configured to determine shoulder multiple stretching information of each match simulation video using a stretching information processing model based on the plurality of match simulation video information of the current match; a strain probability distribution map determination module configured to determine a shoulder strain probability distribution map of each match simulation video based on the shoulder muscle acid swelling degree distribution map and the shoulder multiple stretching information of each match simulation video; a strain risk value determination module configured to determine a shoulder strain risk value of the current match based on a graph neural network for the shoulder strain probability distribution map of each match simulation video; and an early warning module configured to send a risk early warning signal to a control terminal if the shoulder strain risk value of the current match is greater than a preset threshold.

[0010] In a possible implementation, the strain risk value determination module is further configured to: construct a strain analysis graph, the strain analysis graph comprising a plurality of match simulation nodes and a plurality of edges between the plurality of match simulation nodes, node features of each match simulation node comprising each match simulation video information, the shoulder strain probability distribution map of each match simulation video, the shoulder muscle information of the athlete after pre-match training, and the shoulder muscle acid swelling degree distribution map, and the edges between the simulation nodes being a credibility difference value of the match simulation video; and process the strain analysis graph based on a graph neural network to obtain the shoulder strain risk value of the current match.

[0011] In a possible implementation, the strain probability distribution map determination module is further configured to: cluster the shoulder multiple stretching information of each match simulation video to obtain K clusters; determine a plurality of high-risk stretching information based on the K clusters; and determine the shoulder strain probability distribution map of each match simulation video based on the plurality of high-risk stretching information and the shoulder muscle acid swelling degree distribution map.

[0012] In a possible implementation, each match simulation video information of the current match comprises a match simulation video and a credibility of the match simulation video.

[0013] According to a third aspect, embodiments of the present application provide an electronic device, comprising: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method as described above, the method comprising: obtaining historical match video of an athlete against an opponent in a competition, pre-competition training video of the athlete, and shoulder muscle information after the pre-competition training of the athlete; generating a plurality of match simulation video information of the current competition based on the historical match video of the athlete against the opponent in the competition, the pre-competition training video of the athlete, and the shoulder muscle information after the pre-competition training of the athlete; determining a shoulder muscle soreness distribution map based on the shoulder muscle information after the pre-competition training of the athlete; determining shoulder multiple stretching information of each match simulation video using a stretching information processing model based on the plurality of match simulation video information of the current competition; determining a shoulder injury probability distribution map of each match simulation video based on the shoulder muscle soreness distribution map and the shoulder multiple stretching information of each match simulation video; determining a shoulder injury risk value of the current competition based on the shoulder injury probability distribution map of each match simulation video using a graph neural network; and if the shoulder injury risk value of the current competition is greater than a preset threshold, sending a risk warning signal to a control terminal.

[0014] According to a fourth aspect, embodiments of the present application provide a computer readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the image visual-based sports competition risk prediction method as described above, the method comprising: obtaining historical match video of an athlete against an opponent in a competition, pre-competition training video of the athlete, and shoulder muscle information after the pre-competition training of the athlete; generating a plurality of match simulation video information of the current competition based on the historical match video of the athlete against the opponent in the competition, the pre-competition training video of the athlete, and the shoulder muscle information after the pre-competition training of the athlete; determining a shoulder muscle soreness distribution map based on the shoulder muscle information after the pre-competition training of the athlete; determining shoulder multiple stretching information of each match simulation video using a stretching information processing model based on the plurality of match simulation video information of the current competition; determining a shoulder injury probability distribution map of each match simulation video based on the shoulder muscle soreness distribution map and the shoulder multiple stretching information of each match simulation video; determining a shoulder injury risk value of the current competition based on the shoulder injury probability distribution map of each match simulation video using a graph neural network; and if the shoulder injury risk value of the current competition is greater than a preset threshold, sending a risk warning signal to a control terminal.

[0015] The application provides a sports event risk prediction method and system based on image vision, which comprises the following steps: acquiring historical match video of an athlete against an opponent, pre-match training video of the athlete and shoulder muscle information of the athlete after pre-match training; generating a plurality of match simulation video information of the current match based on the historical match video of the athlete against the opponent, the pre-match training video of the athlete and the shoulder muscle information of the athlete after pre-match training; determining a shoulder muscle soreness degree distribution map based on the shoulder muscle information of the athlete after pre-match training; determining shoulder multiple stretching information of each match simulation video based on the plurality of match simulation video information of the current match by using a stretching information processing model; determining a shoulder strain probability distribution map of each match simulation video based on the shoulder muscle soreness degree distribution map and the shoulder multiple stretching information of each match simulation video; determining a shoulder strain risk value of the current match based on the shoulder strain probability distribution map of each match simulation video by using a graph neural network; and if the shoulder strain risk value of the current match is greater than a preset threshold, sending a risk warning signal to a control terminal, so that the sports event shoulder strain risk can be accurately predicted in combination with the action characteristics of the opponent and the state of the athlete. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of a sports event risk prediction method based on image vision is provided for the embodiment of the application.

[0017] Figure 2 A flowchart of determining a shoulder strain probability distribution map of each match simulation video is provided for the embodiment of the application.

[0018] Figure 3 A flowchart of determining a shoulder strain risk value of the current match is provided for the embodiment of the application.

[0019] Figure 4 A schematic diagram of a control terminal is provided for the embodiment of the application.

[0020] Figure 5 A sports event risk prediction system based on image vision is provided for the embodiment of the application. DETAILED DESCRIPTION

[0021] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0022] In this embodiment of the invention, the following are provided: Figure 1 The image-based vision-based sports event risk prediction method is shown, comprising steps S1 to S7:

[0023] Step S1: Obtain historical head-to-head videos of the athlete's opponents, pre-match training videos of the athlete, and shoulder muscle information after the athlete's pre-match training.

[0024] Athlete competitions can include two-person competitive sports such as badminton and tennis.

[0025] Historical head-to-head videos of athletes' opponents refer to video records of past matches between the athletes who will soon be playing against them.

[0026] Historical match videos of athletes against their opponents can be used to analyze the opponent's tactics and playing habits.

[0027] Historical match videos of athletes against their opponents can include full match footage or clips of the opponent in different events.

[0028] The athletes' pre-competition training video is a video taken during the training process of the athletes before the start of the official competition.

[0029] The pre-match training videos record footage of athletes practicing serving, receiving, and footwork. These videos can be used to capture the athletes' current competitive state and the fluidity of their movements.

[0030] The shoulder muscle information of the athlete after the end of pre-match training is physiological state data of the shoulder muscle group of the athlete after completing pre-match training, which is obtained through a physiological monitoring device. The shoulder muscle information of the athlete after the end of pre-match training can accurately reflect the physiological activation state, fatigue degree and muscle function performance of the athlete's shoulder muscle after pre-match training.

[0031] In step S2, based on the historical match video of the athlete against the opponent, the pre-match training video of the athlete and the shoulder muscle information of the athlete after the end of pre-match training, a plurality of match simulation video information of the current match is generated.

[0032] In some embodiments, the match simulation model can be used to generate a plurality of match simulation video information of the current match. The match simulation model is a gated recurrent unit. The input of the match simulation model is the historical match video of the athlete against the opponent, the pre-match training video of the athlete and the shoulder muscle information of the athlete after the end of pre-match training, and the output of the match simulation model is a plurality of match simulation video information of the current match.

[0033] The gated recurrent unit (GRU) can be used to process sequence data and time series information. The gated recurrent unit can control the flow and forgetting of information through the reset gate and the update gate. The reset gate is used to determine whether to ignore historical information, and the update gate can be used to adjust the fusion ratio of historical information and current input information. The gated recurrent unit can more efficiently capture long-term dependencies in sequence data.

[0034] The plurality of match simulation video information of the current match is information describing the possible confrontation process between the home athlete and the opponent in the upcoming match, which is generated by the match simulation model. Each match simulation video information of the current match contains a match simulation video and the credibility value of the corresponding match simulation video.

[0035] The match simulation video is a video of a simulated match between the home athlete and the opponent in the upcoming match. The match simulation video simulates the ball hitting route, running position movement and tactical game process of the home athlete and the opponent.

[0036] The credibility of the match simulation video is a numerical index for measuring the degree of agreement between the generated match simulation video and the real match situation. The credibility of the match simulation video reflects the probability of the occurrence of the simulated confrontation.

[0037] The historical match video of the athlete event against the opponent can provide the tactical characteristics of the opponent, the pre-match training video of the athlete reflects the current action habits of the athlete, and the shoulder muscle information after the pre-match training of the athlete limits the physiological movement limit of the athlete in the simulation. The opponent behavior pattern in the historical match video of the athlete event against the opponent provides the confrontation logic for the simulation of the model, the pre-match training video of the athlete provides the action basis of the athlete for the simulation of the model, and the shoulder muscle information as a physiological constraint condition can prevent the generation of actions beyond the current physical endurance of the athlete, thereby ensuring the rationality of the content of the simulation video of the match.

[0038] The gating cycle unit has the advantage of processing sequence data, can deeply mine the time sequence information in the historical match video of the athlete event against the opponent and the pre-match training video of the athlete, capture the rules such as action sequence and tactical change contained in both, and the model can simulate multiple match scenes in accordance with the actual situation by combining the physiological state constraint reflected by the shoulder muscle information after the pre-match training of the athlete, forming multiple match simulation videos of the competition. The gating cycle unit can calculate the credibility of each match simulation video by analyzing the integrity, relevance of the input data and the rationality of the simulation process, thereby obtaining multiple match simulation video information of the competition.

[0039] In some embodiments, the generating multiple match simulation video information of the competition based on the historical match video of the athlete event against the opponent, the pre-match training video of the athlete and the shoulder muscle information after the pre-match training of the athlete includes steps S21-S23:

[0040] Step S21, determining multiple opponent attack action segments, multiple opponent defense action segments, occurrence frequency of each action of the opponent and threat score of each action of the opponent based on the historical match video of the athlete event against the opponent.

[0041] In some embodiments, the gating cycle unit can be used to determine multiple opponent attack action segments, multiple opponent defense action segments, occurrence frequency of each action of the opponent and threat score of each action of the opponent.

[0042] The multiple opponent attack action segments are extracted from the historical match video of the athlete event against the opponent by the gating cycle unit, and are composed of multiple independent video segments corresponding to different types of attack behaviors initiated by the opponent, each video segment recording a specific attack action with clear attack intention, execution logic and unique action type.

[0043] The opponent's various defense action segments are extracted from the historical match video of the athlete's competition against the opponent by the gated recurrent unit, and are composed of a set of multiple independent video segments corresponding to different types of defensive behaviors made by the opponent to resist attacks and protect the defensive area. Each video segment records a specific defensive action with a clear defensive direction, execution logic, and unique action type.

[0044] Each attack action and each defense action can occur once or multiple times.

[0045] The occurrence frequency of each action of the opponent is a quantitative indicator of the frequency of occurrence of each type of attack action and each type of defense action of the opponent in historical matches, which is calculated by the gated recurrent unit from the historical match video of the athlete's competition against the opponent. The occurrence frequency of each action of the opponent can be represented by the proportion of the number of occurrences of each type of action in the total number of actions.

[0046] The threat score of each action of the opponent is a quantitative score of the threat level of each type of attack action and each type of defense action of the opponent in the game, which is analyzed by the gated recurrent unit from the historical match video of the athlete's competition against the opponent. The higher the score, the stronger the threat level of the type of action.

[0047] The gated recurrent unit can process the time series data of the historical match video of the athlete's competition against the opponent by resetting the gate and updating the gate. The reset gate can filter key features related to actions in the video frames, such as limb motion trajectories and ball motion directions, and can accurately locate the start and end frames of actions with offensive or defensive intent, thereby forming complete action sequences, and ultimately obtaining the opponent's various attack action segments and the opponent's various defense action segments. The update gate of the model can remember the number of occurrences of different action sequences in the video, and calculate the occurrence proportion of each type of action in combination with the total video duration or total number of actions. The model can also output the threat score of each action of the opponent based on the historical features such as the execution speed and attack range of the remembered action, and in combination with the defense pressure data caused to the athlete by the type of action in the historical match video of the athlete's competition against the opponent.

[0048] In step S22, based on the pre-match training video of the athlete and the shoulder muscle information after the end of the pre-match training of the athlete, a set of athlete training advantage action segments, a set of athlete training short board action segments, an athlete shoulder bearable action type, and an athlete shoulder maximum activity angle are determined.

[0049] In some embodiments, a gated recurrent unit can be used to determine a set of athlete training advantage action segments, a set of athlete training short board action segments, an athlete shoulder bearable action type, and an athlete shoulder maximum activity angle.

[0050] The athlete training advantage action fragment set is extracted from the athlete pre-match training video by the gated recurrent unit, and is a set composed of multiple independent video fragments corresponding to different types of actions performed by the home athlete with excellent performance. Each video fragment records a specific training action with high execution success rate, strong action specification, optimal completion efficiency, and unique action type.

[0051] The athlete training short board action fragment set is extracted from the athlete pre-match training video by the gated recurrent unit, and is a set composed of multiple independent video fragments corresponding to different types of actions performed by the home athlete with poor performance. Each video fragment records a specific training action with low execution success rate, weak action specification, poor completion efficiency, and unique action type.

[0052] The athlete shoulder bearable action type is a set of action categories determined by the gated recurrent unit that the shoulder muscles and joints of the home athlete can safely bear without excessive damage risk.

[0053] The athlete shoulder maximum activity angle is the limit angle value that the shoulder joint of the home athlete can reach in each activity direction (forward flexion, backward extension, abduction, adduction, rotation) determined by the gated recurrent unit.

[0054] The gated recurrent unit is good at capturing time series data association. The gated recurrent unit can filter key information such as action execution details and physiological characteristics in muscle electrical signals in the athlete pre-match training video through the reset gate, and can accurately distinguish between high success rate, strong specification advantage actions and low success rate, weak specification short board actions, thereby determining the corresponding action fragment set. The gated recurrent unit can determine the action type that the shoulder can safely bear by updating the gate to remember the physiological data such as muscle strength and endurance and the association rule of action execution, and determine the athlete's shoulder maximum activity angle in combination with the joint activity timing characteristics.

[0055] In step S23, based on the opponent's multiple attack action fragments, the opponent's multiple defense action fragments, the opponent's action frequency, the opponent's action threat score, the athlete training advantage action fragment set, the athlete training short board action fragment set, the athlete shoulder bearable action type, and the athlete shoulder maximum activity angle, multiple exchange simulation video information of the current competition is determined.

[0056] In some embodiments, the gated recurrent unit can be used to determine the multiple exchange simulation video information of the current competition.

[0057] The gating cycle unit has the ability of timing correlation modeling and multi-source information fusion. The gating cycle unit can mine the internal correlation of technical features, occurrence frequency and threat score of various action fragments of the opponent through the reset gate, and can also associate the execution characteristics of the player's training advantages and short board actions, so as to filter out the action confrontation scenes with high probability in the competition. The update gate can remember the physiological boundary of the action type and the maximum activity angle that the shoulder can withstand, and can also perform action adaptation and intensity regulation on the filtered confrontation scenes, so that the gating cycle unit can simulate the sparring process that meets the technical characteristics of both parties and the physiological tolerance of the player, and finally generate multiple sparring simulation video information close to the actual competition situation.

[0058] In step S3, a shoulder muscle soreness distribution map is determined based on the shoulder muscle information after the pre-competition training of the player.

[0059] In some embodiments, a muscle state analysis model can be used to determine the shoulder muscle soreness distribution map. The muscle state analysis model is a Transformer model. The input of the muscle state analysis model is the shoulder muscle information after the pre-competition training of the player, and the output of the muscle state analysis model is the shoulder muscle soreness distribution map.

[0060] The Transformer model includes an encoder and a decoder. The role of the encoder is to perform representation learning on the input sequence, which includes a self-attention mechanism and a feed-forward neural network. The decoder additionally introduces a multi-head attention mechanism based on the encoder, and is used to decode the encoder output and generate the target sequence. The Transformer model can process sequence data in parallel and focus on the correlation between different positions in the input data.

[0061] The shoulder muscle soreness distribution map is a spatial distribution image that can intuitively show the muscle soreness degree of each region of the player's shoulder output by the muscle state analysis model. The shoulder muscle soreness distribution map can distinguish the physiological state of the muscle soreness degree of different muscle parts by color depth and numerical annotation.

[0062] The strength of the electromyographic signal contained in the shoulder muscle information after the pre-competition training of the player is directly related to the degree of muscle fatigue. High-frequency high-amplitude electromyographic signals often correspond to the muscle tension state, and the frequency spectrum change of the signal reflects the accumulation of metabolic products. These microscopic physiological electrical signal data provide the basis for the model to convert from the electromyographic signal domain to the shoulder muscle soreness distribution map.

[0063] The self-attention mechanism of the Transformer model can comprehensively extract features from the shoulder muscle information of the athletes after the pre-match training, capture the correlation and distribution rules between different muscle region electrical signal data, and the model can focus on the spatial correlation between different signal collection points and the trend of signal change over time. The encoder can map the discrete electromyographic signal into a high-dimensional feature vector, and the decoder can convert the feature vector into a two-dimensional distribution expression of the corresponding shoulder anatomy. Through the multi-head attention mechanism, the Transformer model can accurately locate the muscle bundle position of the signal anomaly, calculate the specific soreness value according to the signal intensity, and finally generate a shoulder muscle soreness distribution map reflecting the overall physiological state of the shoulder.

[0064] In step S4, based on the plurality of simulation video information of the match, a stretching information processing model is used to determine the shoulder multiple stretching information of each simulation video.

[0065] The stretching information processing model is a recurrent neural network. The input of the stretching information processing model is the plurality of simulation video information of the match, and the output of the stretching information processing model is the shoulder multiple stretching information of each simulation video.

[0066] A recurrent neural network (RNN) is a neural network model specially used for processing sequence data. The output of the hidden layer in the recurrent neural network is fed back to the input end and participates in the calculation of the next time step. Through this circular structure, the recurrent neural network can remember the information of the previous time step, so as to capture the time sequence dependence in the sequence data. The recurrent neural network is suitable for processing video frame sequences with sequential information.

[0067] The shoulder multiple stretching information of the simulation video is the information output by the stretching information processing model, which reflects the specific details of each stretching action of the athlete's shoulder during the simulated match. Each stretching information includes stretching angle, stretching speed, stretching duration, stretching action starting position, stretching action ending position, muscle contraction mode during stretching, and stretching action continuity level.

[0068] The muscle contraction mode during stretching refers to the specific form of muscle contraction or relaxation when the athlete's shoulder performs the stretching action, such as active stretching, passive stretching, and explosive stretching.

[0069] The stretching action continuity level is a numerical indicator of the smoothness of the connection between the athlete's multiple stretching actions, which can reflect whether the transition between adjacent stretching actions is smooth or not.

[0070] Each of the multiple interaction simulation video information of the competition presents a simulated interaction scene, which contains various sequences of movements of the shoulders of the athletes. The interaction simulation video, as a visual input, can provide the model with the time and space information required to determine the type of movement.

[0071] The recurrent neural network can scan and analyze each frame of image of each interaction simulation video. Through the memory function of the recurrent unit, the recurrent neural network can track the position changes of the shoulder joint key points of the athletes in the continuous frames. When the shoulder joint angle exceeds the threshold of normal smooth movement or a large swing movement is detected, the recurrent neural network can identify and record the movement characteristics of the time period. The recurrent neural network can calculate the starting point and ending point of the movement by integrating the context information of the previous and subsequent frames, thereby quantifying the specific parameters of the stretch, and finally determining the shoulder stretch information of each interaction simulation video.

[0072] Step S5, determining a shoulder strain probability distribution map of each interaction simulation video based on the shoulder muscle soreness degree distribution map and the shoulder stretch information of each interaction simulation video.

[0073] In some embodiments, Figure 2 A flowchart for determining a shoulder strain probability distribution map of each interaction simulation video is provided for the embodiments of the present application. The determination of the shoulder strain probability distribution map of each interaction simulation video includes steps S51-S53:

[0074] Step S51, clustering based on the shoulder stretch information of each interaction simulation video to obtain K clusters.

[0075] The clustering is a K-means clustering algorithm, which is an unsupervised learning algorithm and can divide the input data set into K discrete clusters according to the similarity of data characteristics. The K value can be preset by humans in advance. The K-means clustering algorithm can calculate the distance between data points and cluster centers as a similarity measurement basis, and constantly update the cluster center positions until the similarity of data points within the cluster is maximized and the similarity of data points between clusters is minimized, and finally obtain a stable clustering result.

[0076] The K clusters are K sets formed by grouping the shoulder stretch information of each interaction simulation video according to similarity. Each cluster contains shoulder stretch information with high similarity in core features such as stretch angle, stretch strength, and stretch speed, and can represent a type of stretch action with similar characteristics. The core features of the shoulder stretch information in different clusters have obvious differences, and different clusters correspond to different types of stretch action patterns.

[0077] The execution process of clustering the shoulder multiple stretching information of each sparring simulation video into K clusters using the K-means clustering algorithm is as follows: first, randomly select K data samples from the shoulder multiple stretching information of each sparring simulation video as initial cluster centers, then calculate the similarity distance between each remaining shoulder multiple stretching information sample and each initial cluster center, and assign each sample to the cluster where the nearest cluster center is located. Then, according to all sample data in each cluster, the center position of each cluster is recalculated, and the cluster center is updated. Repeat the steps of sample assignment and cluster center update until the cluster center position no longer changes significantly or the preset number of iterations is reached, and then the clustering process is terminated, thereby obtaining K stable clusters.

[0078] Clustering into K clusters can classify and organize the massive and dispersed shoulder multiple stretching information according to feature similarity, and can reduce data dimension and analysis complexity. The stretching information features in the same cluster are uniform, and similar stretching actions are concentrated and classified, so that the data can be more regular. The feature differences between different clusters are clear, and different clusters can clearly distinguish different types of stretching action patterns to avoid feature confusion caused by the random distribution of various stretching information, thereby providing a clear and orderly data basis for further analysis.

[0079] In step S52, a plurality of high-risk stretching information is determined based on the K clusters.

[0080] In some embodiments, a dangerous stretching analysis model can be used to determine the plurality of high-risk stretching information. The dangerous stretching analysis model is a Transformer model. The input of the dangerous stretching analysis model is the K clusters, and the output of the dangerous stretching analysis model is the plurality of high-risk stretching information.

[0081] The plurality of high-risk stretching information is the specific description information of extreme or non-standard stretching actions that are easy to cause injuries to athletes, which is determined by the dangerous stretching analysis model. The plurality of high-risk stretching information includes high-risk action types, high-risk action occurrence frequencies, high-risk action corresponding force directions, and high-risk action muscle pulling durations.

[0082] The high-risk action types include shoulder over-external stretching, shoulder sharp rotation backward stretching, shoulder over-physiological range forward bending stretching, shoulder explosive stretching in weight-bearing state, and shoulder asymmetric twisting stretching.

[0083] The high-risk action occurrence frequency refers to the number of times of occurrence of each type of high-risk stretching action in the sparring scene corresponding to a single sparring simulation video.

[0084] The high-risk action corresponding force direction refers to the specific direction of force when the high-risk stretching action acts on the shoulder muscles, such as longitudinal pulling, transverse tearing, and oblique twisting.

[0085] The muscle pull duration of a high-risk action refers to the time length during which a single high-risk stretching action continuously acts on the shoulder muscles.

[0086] The K clusters organize the originally disordered stretching data into statistically meaningful categories, and make the data exhibit a structured characteristic distribution. Different clusters represent different intensity of movement patterns, which contain abnormal data groups deviating from the normal physiological range. These clustered data sets highlight the characteristics of extreme actions, thereby providing a clear classification basis for identifying potential risks.

[0087] The Transformer model can deeply analyze the internal features of each cluster in the K clusters through the global attention mechanism. The Transformer model can compare the stretching amplitude, duration, and other parameters within the cluster with known movement injury mechanics models. The Transformer model can extract cluster center features through the encoder and identify stretching patterns that exceed the safe bearing range of human joints. The decoder converts these dangerous patterns into specific textual or numerical descriptions, thereby filtering and determining multiple high-risk stretching information.

[0088] In step S53, a shoulder injury probability distribution map of each sparring simulation video is determined based on the multiple high-risk stretching information and the shoulder muscle soreness degree distribution map.

[0089] In some embodiments, a shoulder injury probability distribution map of each sparring simulation video can be determined using an injury probability determination model. The injury probability determination model is a Transformer model. The input of the injury probability determination model is the multiple high-risk stretching information and the shoulder muscle soreness degree distribution map, and the output of the injury probability determination model is the shoulder injury probability distribution map of each sparring simulation video.

[0090] The shoulder injury probability distribution map of each sparring simulation video is a spatial distribution image of the possibility of shoulder injury in the simulated competition process output by the injury probability determination model.

[0091] The multiple high-risk stretching information clearly identifies the characteristics of high-risk stretching actions, which can directly point to the key action factors that may cause shoulder injuries, thereby providing specific basis for judging the damage risk of different high-risk stretching actions to the shoulder muscles. The shoulder muscle soreness degree distribution map shows the basic state of soreness in each region of the shoulder, and the degree of soreness can directly reflect the muscle tolerance.

[0092] The transformer model can process multiple high-risk stretching information and shoulder muscle soreness distribution maps at the same time. The transformer model can capture the correlation between the risk features in the high-risk stretching information and the acidification states of each region in the shoulder muscle soreness distribution map through the self-attention mechanism. The model can analyze the load influence of each high-risk stretching action on different regions of the shoulder, and then combine the acidification basic state of the corresponding region in the shoulder muscle soreness distribution map to quantitatively calculate the probability of injury of each region in the hand-to-hand simulation video scene. The model can map this probability back to the anatomical structure diagram of the shoulder, and finally generate a shoulder injury probability distribution map of each hand-to-hand simulation video that can reflect the dynamic risk.

[0093] Step S6, determining the shoulder injury risk value of the current competition based on the graph neural network of the shoulder injury probability distribution map of each hand-to-hand simulation video.

[0094] In some embodiments, Figure 3 A flowchart for determining the shoulder injury risk value of the current competition is provided for the embodiments of the present application. The determination of the shoulder injury risk value of the current competition includes steps S61-S62:

[0095] Step S61, constructing an injury analysis graph, the injury analysis graph including a plurality of hand-to-hand simulation nodes and a plurality of edges between the plurality of hand-to-hand simulation nodes, the node features of each hand-to-hand simulation node including each hand-to-hand simulation video information, the shoulder injury probability distribution map of each hand-to-hand simulation video, the shoulder muscle information after the pre-competition training of the athlete, and the shoulder muscle soreness distribution map, and the edges between the hand-to-hand simulation nodes being the credibility difference values of the hand-to-hand simulation videos.

[0096] The injury analysis graph is a graph data structure that can describe the correlation between different simulation competition scenarios and the comprehensive risk features. The injury analysis graph includes a plurality of hand-to-hand simulation nodes and a plurality of edges between the plurality of hand-to-hand simulation nodes, and organizes all simulation data and risk assessment results in a structured form of graph.

[0097] The plurality of hand-to-hand simulation nodes is the core component of the injury analysis graph, each node corresponding to a hand-to-hand simulation scenario, and the edges between the hand-to-hand simulation nodes being the credibility difference values of the hand-to-hand simulation videos associated with the two hand-to-hand simulation nodes.

[0098] Step S62, processing the injury analysis graph based on the graph neural network to obtain the shoulder injury risk value of the current competition.

[0099] A graph neural network (GNN) is a deep learning model that can run directly on a graph. The graph neural network can update the feature representation of a node through a neighborhood aggregation mechanism, i.e., a message passing mechanism. Each node in a graph can collect and aggregate information of its neighbor nodes, thereby fusing local structure and global context information. The graph neural network can process data in a non-Euclidean space and effectively capture the dependency relationship and topological structure characteristics between nodes. The input of the graph neural network is the strain analysis graph, and the output of the graph neural network is the shoulder strain risk value of the current match.

[0100] The shoulder strain risk value of the current match is a numerical index determined by the graph neural network to quantitatively evaluate the risk degree of the shoulder strain of the home athlete in the current match.

[0101] The shoulder strain risk value of the current match is a numerical value within a specific interval. The higher the risk value, the greater the possibility of shoulder muscle strain of the athlete in the current match, and the closer the sports load of the shoulder muscle to the injury critical state.

[0102] The strain analysis graph can represent all information related to the shoulder strain risk of the current match in the form of a graph, and can intuitively present multiple match simulation scenarios and their associated relationships. Each match simulation node represents the required features for strain risk analysis in the corresponding scenario, including each match simulation video information, the shoulder strain probability distribution map of each match simulation video, the shoulder muscle information of the athlete after the pre-match training, and the shoulder muscle soreness distribution map. The edges between nodes are characterized by the credibility difference of the match simulation videos associated with the two match simulation nodes. This structured representation method facilitates the understanding and processing of complex relationships between scenarios by the graph neural network.

[0103] By the collaborative support of node features and edges, the graph neural network can more accurately capture the risk feature differences of different sparring simulation scenarios and the mutual influence caused by the credibility differences between scenarios. The shoulder injury probability distribution in the node feature reflects the risk distribution of different areas of the shoulder in the corresponding scenario, and the credibility difference on the edge reflects the reliability degree correlation of risk data in different scenarios. The construction of the injury analysis graph can effectively organize the scattered risk-related data and scenario correlation, providing a clear data structure for the graph neural network processing. This structured organization method can also reduce the complexity of data processing, thereby improving the training and inference efficiency of the model. At the same time, with the help of the structured node and edge information provided by the injury analysis graph, the graph neural network model can more efficiently process multi-scenario risk data, avoiding the data sparsity or dimension redundancy problems that may occur in traditional risk assessment methods. Relying on the structured advantages of the graph, the graph neural network is more targeted when dealing with complex correlations between scenarios. By comprehensively analyzing the features and correlation of all sparring simulation nodes, the graph neural network can more accurately and comprehensively integrate multi-scenario risk information. Finally, the graph neural network can accurately calculate the risk value that objectively reflects the shoulder injury possibility of this competition.

[0104] The graph neural network can perform deep processing on the injury analysis graph. The graph neural network can mine the correlation between various types of data within each sparring simulation node by aggregating the feature information of each node. Meanwhile, the graph neural network can use the credibility difference information represented by the edges between nodes to weigh the weights of different sparring simulation nodes in risk assessment. Through the multi-layer network structure, the graph neural network extracts and fuses the feature information and correlation in the graph layer by layer, and comprehensively considers the injury risk features in all sparring simulation scenarios and the credibility weights of each scenario. Finally, the graph neural network can calculate a quantitative value that can comprehensively reflect the risk degree of shoulder injury in this competition, i.e., the shoulder injury risk value of this competition.

[0105] In step S7, if the shoulder injury risk value of this competition is greater than a preset threshold, a risk warning signal is sent to a control terminal.

[0106] When the shoulder injury risk value of this competition is determined, the shoulder injury risk value of this competition is judged. If the shoulder injury risk value of this competition is greater than a preset threshold, a risk warning signal is sent to a control terminal to remind the athlete to participate in the competition carefully. Figure 4 A schematic diagram of a control terminal provided for an embodiment of the present application is shown.

[0107] Based on the same inventive concept, Figure 5 A schematic diagram of a sports event risk prediction system based on image vision provided for an embodiment of the present application is shown. The sports event risk prediction system based on image vision includes:

[0108] The acquisition module 81 is configured to acquire historical match video of the athlete against an opponent, pre-match training video of the athlete, and shoulder muscle information after the pre-match training of the athlete ends;

[0109] The simulation module 82 is configured to generate a plurality of match simulation video information of the current match based on the historical match video of the athlete against the opponent, the pre-match training video of the athlete, and the shoulder muscle information after the pre-match training of the athlete ends.

[0110] The acid swelling degree distribution map determination module 83 is configured to determine a shoulder muscle acid swelling degree distribution map based on the shoulder muscle information after the pre-match training of the athlete ends.

[0111] The stretching information determination module 84 is configured to determine shoulder multiple stretching information of each match simulation video based on the plurality of match simulation video information of the current match using a stretching information processing model.

[0112] The strain probability distribution map determination module 85 is configured to determine a shoulder strain probability distribution map of each match simulation video based on the shoulder muscle acid swelling degree distribution map and the shoulder multiple stretching information of each match simulation video.

[0113] The strain risk value determination module 86 is configured to determine a shoulder strain risk value of the current match based on a graph neural network for the shoulder strain probability distribution map of each match simulation video.

[0114] The warning module 87 is configured to send a risk warning signal to a control terminal if the shoulder strain risk value of the current match is greater than a preset threshold.

[0115] It should be noted that, in order to simplify the expression disclosed in the present specification, and to help the understanding of one or more embodiments of the present application, in the foregoing description of the embodiments of the present specification, various features are sometimes combined into one embodiment, figure or description thereof. However, this disclosure method does not mean that the features required by the present specification are more than the features mentioned in the claims. In fact, the features of the embodiments are less than all the features of the disclosed single embodiment.

[0116] Finally, it should be understood that the embodiments described in the present specification are only used to illustrate the principles of the embodiments of the present specification. Other variations can also belong to the scope of the present specification. Therefore, as an example but not limitation, alternative configurations of the embodiments of the present specification can be considered consistent with the teachings of the present specification. Accordingly, the embodiments of the present specification are not limited to the embodiments explicitly introduced and described in the present specification.

Claims

1. A method for predicting the risk of sports events based on image vision, characterized in that, include: Obtain historical head-to-head videos of athletes' opponents, pre-match training videos of athletes, and shoulder muscle information of athletes after pre-match training; Based on the athlete's historical head-to-head videos against opponents, the athlete's pre-match training videos, and the athlete's shoulder muscle information after pre-match training, multiple simulated head-to-head video information for this match is generated. A distribution map of shoulder muscle soreness was determined based on the shoulder muscle information of the athletes after their pre-competition training. Based on the simulated video information of multiple matches in this competition, the stretching information processing model was used to determine the multiple shoulder stretching information of each simulated match video. Based on the shoulder muscle soreness distribution map and the shoulder stretching information of each sparring simulation video, a shoulder strain probability distribution map for each sparring simulation video is determined. The shoulder strain risk value for this match is determined based on the probability distribution map of shoulder strain in each simulated match video using a graph neural network. If the risk of shoulder strain in this competition exceeds a preset threshold, a risk warning signal will be sent to the control terminal.

2. The sports event risk prediction method based on image vision as described in claim 1, characterized in that, The shoulder strain risk value for this match is determined by analyzing the shoulder strain probability distribution map of each simulated match video based on a graph neural network, including: A muscle strain analysis atlas is constructed, which includes multiple simulated competition nodes and multiple edges between the simulated competition nodes. The node features of each simulated competition node include information from each simulated competition video, a probability distribution map of shoulder strain in each simulated competition video, shoulder muscle information of the athlete after pre-competition training, and a distribution map of shoulder muscle soreness. The edges between the simulated competition nodes represent the credibility difference between the simulated competition videos. The shoulder strain risk value for this competition was obtained by processing the strain analysis map using a graph neural network.

3. The sports event risk prediction method based on image vision as described in claim 1, characterized in that, The determination of the shoulder strain probability distribution map for each sparring simulation video based on the shoulder muscle soreness distribution map and the shoulder stretching information from each sparring simulation video includes: K clusters were obtained by clustering the shoulder stretching information from each simulated engagement video. Multiple high-risk stretching information is determined based on the K clusters; Based on the multiple high-risk stretching information and the shoulder muscle soreness distribution map, a shoulder strain probability distribution map is determined for each simulated combat video.

4. The sports event risk prediction method based on image vision as described in claim 1, characterized in that, The information for each simulated match in this competition includes the simulated match video itself and the credibility of the simulated match video.

5. A sports event risk prediction system based on image vision, characterized in that, include: The acquisition module is used to acquire historical head-to-head videos of athletes' opponents in competitions, pre-competition training videos of athletes, and shoulder muscle information of athletes after pre-competition training. The simulation module is used to generate multiple simulated match videos for this competition based on the athlete's historical head-to-head videos against opponents, the athlete's pre-match training videos, and the athlete's shoulder muscle information after the pre-match training. The soreness distribution map determination module is used to determine the soreness distribution map of the shoulder muscles based on the shoulder muscle information of the athlete after the pre-competition training. The stretching information determination module is used to determine the multiple shoulder stretching information in each simulated match video based on the multiple match simulation video information of this competition using the stretching information processing model; The strain probability distribution map determination module is used to determine the shoulder strain probability distribution map for each sparring simulation video based on the shoulder muscle soreness distribution map and the shoulder stretching information of each sparring simulation video. The shoulder strain risk determination module is used to determine the shoulder strain risk value for this match based on the probability distribution map of shoulder strain in each simulated match video using a graph neural network. The early warning module is used to send a risk warning signal to the control terminal if the shoulder strain risk value in this competition is greater than a preset threshold.

6. The sports event risk prediction system based on image vision as described in claim 5, characterized in that, The strain risk value determination module is also used for: A muscle strain analysis atlas is constructed, which includes multiple simulated competition nodes and multiple edges between the simulated competition nodes. The node features of each simulated competition node include information from each simulated competition video, a probability distribution map of shoulder strain in each simulated competition video, shoulder muscle information of the athlete after pre-competition training, and a distribution map of shoulder muscle soreness. The edges between the simulated competition nodes represent the credibility difference between the simulated competition videos. The shoulder strain risk value for this competition was obtained by processing the strain analysis map using a graph neural network.

7. The image-vision-based sports event risk prediction system as described in claim 5, characterized in that, The strain probability distribution map determination module is also used for: K clusters were obtained by clustering the shoulder stretching information from each simulated engagement video. Multiple high-risk stretching information is determined based on the K clusters; Based on the multiple high-risk stretching information and the shoulder muscle soreness distribution map, a shoulder strain probability distribution map is determined for each simulated combat video.

8. The image-based sports event risk prediction system as described in claim 5, characterized in that, The information for each simulated match in this competition includes the simulated match video itself and the credibility of the simulated match video.

9. An electronic device, characterized in that, include: processor; Memory; And a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the image vision-based sports event risk prediction method as described in any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the image-based sports event risk prediction method as described in any one of claims 1 to 4.

Images