An AI and image analysis-based sports event training prediction method and system

Abstract

The application provides a sports competition training prediction method and system based on AI and image analysis, and relates to the technical field of sports competition training prediction. The method comprises the following steps: acquiring a video of a badminton match of an athlete; determining a plurality of ball fault points by using a fault point determination model based on the video of the badminton match of the athlete; determining a K value based on the plurality of ball fault points; performing clustering based on the plurality of ball fault points and the K value to determine K fault regions; constructing a fault region atlas; processing the fault region atlas based on a graph neural network to determine at least K hitting training points, wherein each fault region has at least one hitting training point; and determining a target training plan based on the at least K hitting training points. The method can accurately formulate an optimal targeted training scheme according to the badminton match faults of the athlete.

Description

Technical Field

[0001] This invention relates to the field of sports event training prediction technology, specifically to a sports event training prediction method and system based on AI and image analysis. Background Technology

[0002] In competitive badminton, the specificity of an athlete's training is directly related to their match performance, and accurately identifying and controlling their mistakes is the core key to improving training efficiency. Current badminton training largely relies on coaches' subjective observations or participants' post-match experience summaries, requiring extensive and repeated reviews of past matches—a time-consuming, laborious, and inefficient process that lacks in-depth and systematic analysis of actual match error data. Traditional training methods often fail to accurately and comprehensively capture the scattered errors made by athletes during matches, resulting in a lack of precise basis for training plan development, easily leading to blind training and a lack of clear training focus. Especially for the individualized error characteristics of different athletes, existing methods struggle to achieve targeted focus, wasting training resources and failing to quickly address technical weaknesses. Furthermore, existing training programs are mostly fixed templates, unable to be dynamically adjusted based on athletes' match errors, thus failing to meet the precise and personalized training needs of athletes.

[0003] Therefore, how to accurately formulate the best targeted training plan based on the mistakes of athletes in badminton matches is an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem this invention addresses is how to accurately formulate the best targeted training program based on the mistakes made by athletes in badminton matches.

[0005] According to a first aspect, the present invention provides a sports event training prediction method based on AI and image analysis, comprising: acquiring badminton match videos of athletes; determining multiple hitting error points using an error point determination model based on the badminton match videos; determining a K value based on the multiple hitting error points; clustering the multiple hitting error points and the K value to determine K error regions; constructing an error region map, the error region map including K error region nodes and multiple edges between the K error region nodes, wherein the node features of each error region node are multiple hitting error points in the error region, and the edges between error region nodes are the straight-line distances between error regions; processing the error region map based on a graph neural network to determine at least K hitting training points, wherein each error region has at least one hitting training point; and determining a target training plan based on the at least K hitting training points.

[0006] In one possible implementation, determining the target training plan based on the at least K hitting training points includes: determining multiple high-importance hitting training points and multiple medium-importance hitting training points based on the athlete's badminton match video and the at least K hitting training points; generating multiple training plans based on the multiple high-importance hitting training points, the multiple medium-importance hitting training points, and the at least K hitting training points; displaying the multiple training plans on the screen, obtaining the training plan selected by the user, and using the training plan selected by the user as the target training plan.

[0007] In one possible implementation, the error point determination model is a recurrent neural network model.

[0008] In one possible implementation, the step of clustering and determining K error regions based on the multiple ball-hitting error points and the K value includes: using the K-means clustering algorithm to determine K clusters based on the multiple ball-hitting error points and the K value; and connecting the ball-hitting error points at the edges of each cluster to obtain K error regions.

[0009] According to a second aspect, the present invention provides a sports event training prediction system based on AI and image analysis, comprising: an acquisition module for acquiring badminton match videos of athletes; an error point determination module for determining multiple hitting error points based on the badminton match videos using an error point determination model; an error point analysis module for determining a K value based on the multiple hitting error points; a clustering module for clustering based on the multiple hitting error points and the K value to determine K error regions; a graph construction module for constructing an error region graph, the error region graph including K error region nodes and multiple edges between the K error region nodes, wherein the node feature of each error region node is multiple hitting error points in the error region, and the edges between error region nodes are the straight-line distances between error regions; a training point determination module for processing the error region graph based on a graph neural network to determine at least K hitting training points, wherein each error region has at least one hitting training point; and a training plan determination module for determining a target training plan based on the at least K hitting training points.

[0010] In one possible implementation, the training plan determination module is further configured to: determine multiple high-importance hitting training points and multiple medium-importance hitting training points based on the athlete's badminton match video and the at least K hitting training points; generate multiple training plans based on the multiple high-importance hitting training points, the multiple medium-importance hitting training points, and the at least K hitting training points; display the multiple training plans on the screen, obtain the training plan selected by the user, and use the training plan selected by the user as the target training plan.

[0011] In one possible implementation, the error point determination model is a recurrent neural network model.

[0012] In one possible implementation, the clustering module is further configured to: determine K clusters based on the multiple ball-hitting error points and the K value using the K-means clustering algorithm; and connect the ball-hitting error points at the edges of each cluster to obtain K error regions.

[0013] According to a third aspect, embodiments of the present invention provide an electronic device, including: 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 including: acquiring a badminton match video of an athlete; determining multiple hitting errors using an error point determination model based on the badminton match video of the athlete; determining a K value based on the multiple hitting errors; clustering based on the multiple hitting errors and the K value to determine K error regions; constructing an error region map, the error region map including K error region nodes and multiple edges between the K error region nodes, wherein the node feature of each error region node is multiple hitting errors in the error region, and the edges between error region nodes are the straight-line distances between error regions; processing the error region map based on a graph neural network to determine at least K hitting training points, wherein each error region has at least one hitting training point; and determining a target training plan based on the at least K hitting training points.

[0014] According to the fourth aspect, this embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned AI- and image analysis-based sports event training prediction method. The method includes: acquiring badminton match videos of athletes; determining multiple hitting errors using an error point determination model based on the badminton match videos; determining a K value based on the multiple hitting errors; clustering the multiple hitting errors and the K value to determine K error regions; constructing an error region map, the error region map including K error region nodes and multiple edges between the K error region nodes, wherein each error region node's node feature is multiple hitting errors within that error region, and the edges between error region nodes are the straight-line distances between error regions; processing the error region map based on a graph neural network to determine at least K hitting training points, wherein each error region has at least one hitting training point; and determining a target training plan based on the at least K hitting training points.

[0015] This invention provides a method and system for predicting sports event training based on AI and image analysis. The method includes: acquiring badminton match videos of athletes; determining multiple hitting errors using an error point determination model based on the badminton match videos; determining a K-value based on the multiple hitting errors; clustering the multiple hitting errors and the K-value to determine K error regions; constructing an error region map, the error region map including K error region nodes and multiple edges between the K error region nodes, wherein the node feature of each error region node is multiple hitting errors in that error region, and the edges between error region nodes are the straight-line distances between error regions; processing the error region map based on a graph neural network to determine at least K hitting training points, wherein each error region has at least one hitting training point; and determining a target training plan based on the at least K hitting training points. This method can accurately formulate the best targeted training plan based on the athlete's badminton match errors. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a sports event training prediction method based on AI and image analysis, provided as an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of a high-speed industrial camera provided in an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of a mobile high-speed shooting device provided in an embodiment of the present invention;

[0019] Figure 4 This is a schematic diagram illustrating the location of a hitting error point on a badminton court, provided as an embodiment of the present invention.

[0020] Figure 5 This is a schematic diagram of a process for determining K error regions provided in an embodiment of the present invention;

[0021] Figure 6 This is a schematic diagram of a process for determining a target training plan, provided by an embodiment of the present invention.

[0022] Figure 7 This is a schematic diagram of a sports event training prediction system based on AI and image analysis, provided as an embodiment of the present invention. Detailed Implementation

[0023] 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.

[0024] In this embodiment of the invention, the following are provided: Figure 1 The method for predicting sports events based on AI and image analysis is shown, and the method includes steps S1 to S7:

[0025] Step S1: Obtain badminton match videos of athletes.

[0026] The badminton match videos are high-definition, high-frame-rate video footage shot using professional high-speed camera equipment deployed at the venue.

[0027] The camera equipment can include high-speed industrial camera sets permanently deployed at the venue, as well as mobile high-speed shooting equipment equipped with image-stabilized gimbals. All cameras are equipped with a global shutter function to avoid image blurring in high-speed motion scenes and ensure the accuracy of video detail reproduction. Figure 2 This is a schematic diagram of a high-speed industrial camera provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of a mobile high-speed shooting device provided in an embodiment of the present invention.

[0028] Videos of badminton matches can fully present the athlete's hitting motions, positioning, scoring and mistakes, and other match details.

[0029] Step S2: Based on the badminton match video of the athlete, use the error point determination model to determine multiple hitting error points.

[0030] The error point determination model is a recurrent neural network model. The input of the error point determination model is the badminton match video of the athlete, and the output of the error point determination model is multiple hitting errors.

[0031] Recurrent Neural Networks (RNNs) are a type of neural network capable of processing sequential data. RNNs utilize internal memory mechanisms to process temporally sequential input data and can capture temporal dependencies within the data. By introducing recurrent connections into the network, RNNs ensure that the output at the current time step depends not only on the current input but also on the hidden states from previous time steps. RNNs can be used to analyze and process temporal data such as video frame sequences and language text, effectively uncovering dynamic features and patterns within the sequential data.

[0032] Multiple hitting error points are a set of specific locations on the court where athletes made hitting errors during a match, along with the corresponding error event characteristics, identified and extracted by analyzing badminton match videos using an error point determination model. Figure 4 This is a schematic diagram showing the location of a badminton court error point, as provided in an embodiment of the present invention.

[0033] Each point of error corresponds to a three-dimensional spatial coordinate at the time of the error, and includes the timestamp of the error, the type of error, the characteristics of the hitting action corresponding to the error, the player's standing area at the time of the error, and the trajectory parameters of the ball at the time of the error.

[0034] Error types are specific tactical classifications of the nature of errors. Error types include subcategories such as returns into the net, baseline shots out of bounds, sideline shots out of bounds, smash errors, lob shots, and serve return errors.

[0035] The characteristics of a hitting motion describe the physical features of an athlete's limbs and equipment movement at the moment of a mistake. These characteristics include the athlete's swing speed and racket face angle at the time of the mistake.

[0036] The athlete positioning area describes the athlete's specific tactical positioning on the field at the time of the error. The athlete positioning area includes whether the athlete was in the frontcourt, midcourt, or backcourt at the time of the error, and their specific location within the left or right half of the court.

[0037] The shuttlecock's trajectory parameters are data recording the shuttlecock's spatial flight state during a mis-hit. These parameters include the shuttlecock's initial flight direction, flight altitude, and the deviation between the predicted landing point and the actual landing point.

[0038] The video of a badminton match contains a sequence of frames showing the athlete's body movements over time, as well as a sequence of the shuttlecock's flight trajectory in the air. The video data contains dynamic temporal information such as the timing of the athlete's swing, the angle of the shot, the speed of movement, and the landing point of the shuttlecock. The error point determination model can locate the specific location and time of the error by analyzing the frame sequence information in the video.

[0039] Recurrent Neural Networks (RNNs), through their unique recurrent structure, can perform frame-by-frame scanning and temporal analysis of badminton match videos. The hidden layer states within the RNN are continuously updated with the continuous input of video frames, thus memorizing and associating continuous features such as the athlete's actions and shuttlecock trajectories at different moments in the badminton match video. The model can learn standard high-level badminton action patterns and shuttlecock trajectories. Then, when processing badminton match videos, it can extract in real time the athlete's swing, footwork, hitting angle, and trajectory features such as the shuttlecock's flight direction and speed in the current frame, and calculate the deviation values ​​of these features from the baseline patterns. When the RNN detects abnormal temporal patterns such as slow footwork causing a deviation in receiving position, deformed swing causing the return to hit the net or go out of bounds, or improper control of the hitting power causing the shuttlecock's flight trajectory to deviate from the expected direction, it can locate the specific location of the shuttlecock landing or the violation at the moment of the violation based on the spatial mapping relationship of these abnormal patterns in the video frames and the coordinate system of the court. At the same time, it extracts and integrates various related features corresponding to the error, and then marks it as a point of error. The model can continuously traverse the entire sequence of match videos, analyze and mark each frame that matches the characteristics of a mistake, and finally integrate all the marked results to output multiple points of mistake in the game.

[0040] In some embodiments, determining multiple hitting errors based on the athlete's badminton match video includes steps S21-S23:

[0041] Step S21: Based on the badminton match video of the athlete, determine the set of timing parameters of multiple single-shot return of the athlete's actions, the set of real-time flight trajectory sequences of multiple single-shot return of the badminton shuttlecock, and the video frames corresponding to abnormal actions.

[0042] In some embodiments, a recurrent neural network can be used to determine a set of timing parameters of multiple single-shot returns of an athlete's actions, a set of real-time flight trajectory sequences of multiple single-shot returns of a badminton shuttlecock, and video frames corresponding to abnormal actions.

[0043] The temporal parameters of the athlete's movement for each single ball return are output by a recurrent neural network. These parameters represent continuous data on the changes in the motion state of various parts of the athlete's body over time during each complete ball return. The temporal parameters for a single ball return include sequences of joint angle changes over time, limb position changes over time, velocity changes over time, and acceleration changes over time.

[0044] The real-time flight trajectory sequence of the shuttlecock for each single return is output by a recurrent neural network. During each complete return of the shuttlecock, the three-dimensional coordinate data of the shuttlecock's spatial position continuously changes over time from being hit to being hit again, landing, or going out of bounds.

[0045] The video frames corresponding to abnormal actions are determined by a recurrent neural network. The video of an athlete's badminton match presents all the frames at the moment when the athlete makes a mistake or does something that does not conform to the normal pattern. The frames can record the body posture information of the athlete when making the abnormal action and the spatial position information of the badminton shuttlecock at that moment.

[0046] Recurrent neural networks (RNNs) possess the ability to model sequences of temporal data and capture spatial features and temporal correlations in videos. They can accurately identify pixel changes, key point correlations in human posture, and trajectory patterns of objects within consecutive frames of a badminton match video. RNNs can extract temporal features of athlete limb movements from consecutive frames of a badminton match video and integrate them into continuous data on the movement states of various body parts during each single shuttlecock return. This forms multiple sets of temporal parameters for athlete actions during single shuttlecock returns. The model can identify the pixel features of the shuttlecock in the video and fit its temporal spatial position changes, thereby generating multiple sets of real-time shuttlecock flight trajectory sequences for single shuttlecock returns. The model can also learn the characteristic patterns of normal hitting actions and label frame sequences in the video that deviate from these patterns, thus identifying the video frames corresponding to abnormal actions.

[0047] Step S22: Based on the set of athlete action timing parameters for multiple single ball returns, the set of badminton shuttlecock real-time flight trajectory sequences for multiple single ball returns, the video frames corresponding to the abnormal actions, determine the athlete's center of gravity adjustment frequency time domain value, the athlete's swing angular velocity change curve, the athlete's movement start speed time history data, and the single-shot action scoring table.

[0048] In some embodiments, recurrent neural networks can be used to determine the time-domain value of the athlete's center of gravity adjustment frequency, the athlete's swing angular velocity change curve, the athlete's movement start speed time history data, and the single-shot hitting action scoring table.

[0049] Center of gravity adjustment frequency refers to the number of times an athlete's center of gravity completes an effective position or posture adjustment per unit of time.

[0050] The time-domain value of the athlete's center of gravity adjustment frequency is a continuous data sequence of the athlete's body center of gravity adjustment frequency changing over time, output by a recurrent neural network.

[0051] The athlete's swing angular velocity change curve is a continuous curve output by a recurrent neural network showing the instantaneous angular velocity of the athlete's racket-holding hand changing with time during the swing process.

[0052] The athlete's starting speed time-course data is the dynamic data of the change of the initial speed of the body or key parts over time when the athlete starts moving from a standstill after receiving the ball signal, which is output by a recurrent neural network.

[0053] The Single-Race Shot Scoring Table is a data table that uses a recurrent neural network to quantitatively evaluate the quality of all single-shot returns in a badminton match. The row dimension of the Single-Race Shot Scoring Table is based on each single-shot return as an independent row unit, with each row corresponding to a complete return action in the match. The column dimension of the Single-Race Shot Scoring Table includes basic information columns, scoring item columns, score result columns, and rating result columns.

[0054] The basic information column includes the timestamp of a single return action and the round number of the shot.

[0055] The scoring criteria include indicators of movement standardization, movement fluency, and ball-hitting speed.

[0056] The score results column records the specific score for each scoring item and the overall score result.

[0057] The rating results are presented in four levels: Excellent, Good, Satisfactory, and Unsatisfactory, based on the comprehensive score.

[0058] Recurrent neural networks (RNNs) can extract temporal variation patterns of motion features such as the athlete's center of gravity adjustment frequency sequence and the athlete's swing angular velocity change curve from multiple sets of temporal parameters of the athlete's single-shot return. This allows them to generate corresponding feature sequences. Simultaneously, RNNs can combine the spatial features of multiple sets of real-time badminton shuttlecock flight trajectory sequences from multiple single-shot returns to establish a correlation between action and trajectory. Furthermore, RNNs can combine the visual features of video frames corresponding to abnormal actions with the quantitative features of multiple sets of temporal parameters of the athlete's single-shot return, thereby learning the feature patterns of abnormal actions and ultimately outputting quantitative analysis results such as the athlete's center of gravity adjustment frequency sequence and a single-shot action scoring table.

[0059] Step S23: Based on the set of real-time flight trajectory sequences of the badminton shuttlecocks from multiple single-shot returns, the time domain value of the athlete's center of gravity adjustment frequency, the change curve of the athlete's swing angular velocity, the time history data of the athlete's movement start speed, and the single-shot hitting action scoring table, determine multiple hitting error points.

[0060] In some embodiments, a recurrent neural network can be used to determine multiple points of misplay.

[0061] Recurrent neural networks (RNNs) can extract key features from multiple sets of real-time badminton trajectory sequences of single-shot returns, time-domain values ​​of athlete's center of gravity adjustment frequency, athlete's swing angular velocity variation curves, athlete's movement start speed time-history data, and single-shot action scoring tables. The model can extract features such as landing point deviation and trajectory stability from the real-time badminton trajectory sequence set; average and peak frequency features from the center of gravity adjustment frequency time-domain values; maximum and angular velocity rise rate features from the swing angular velocity variation curves; and start time and average start speed features from the movement start speed time-history data. It also treats each score in the single-shot action scoring table as an independent feature. The model can use temporal correlation analysis techniques to align and correlate features corresponding to the same hitting action across different dimensions, and then construct feature chains to analyze the relationships between these features. Finally, the model can judge the comprehensive characteristics of each shot based on the preset error judgment rules and thresholds. When the characteristics of a shot meet the error judgment conditions, it is determined as a shot error point. After analyzing and judging each shot one by one, the model can finally determine multiple shot error points.

[0062] Step S3: Determine the K value based on the multiple ball-hitting error points.

[0063] In some embodiments, a fault analysis model can be used to determine the K value. The fault analysis model is a deep neural network model. The input to the fault analysis model is the plurality of hitting errors, and the output of the fault analysis model is the K value.

[0064] Deep neural network models include deep neural networks (DNNs), which are neural networks based on deep learning architectures and can contain multiple hidden layers. Deep neural networks can abstract features from input data and map them layer by layer through multiple layers of nonlinear transformations. They can learn the complex spatial distribution patterns of data and adjust inter-layer weights through backpropagation, thus possessing powerful function fitting and pattern recognition capabilities.

[0065] The K value is the optimal number of clusters for clustering, calculated by comprehensively evaluating the distribution characteristics of multiple hitting errors using an error point analysis model.

[0066] Multiple hitting errors were recorded, containing multi-dimensional attribute data including three-dimensional spatial coordinates, error type, hitting action characteristics, and ball trajectory parameters. This data not only reflects the degree of clustering of errors in physical space but also demonstrates the inherent correlation between errors in their causes, action patterns, and tactical scenarios. Multiple hitting errors can reveal the concentration trends and distribution patterns of athlete errors, and the error point analysis model can determine the appropriate number of clusters through in-depth analysis of these features.

[0067] Deep neural networks can analyze the spatial distribution density of hitting errors, identifying densely populated areas of high error rates on the court. Simultaneously, the model can combine error type and hitting motion characteristics to uncover the similarities in motion patterns hidden behind spatial locations. For example, the model can detect that errors near the baseline are often related to excessive swing speed, while errors at the net are often related to improper racket face angle control. Through cross-combination of these features and deep feature extraction, deep neural networks can assess how many groups to divide the overall errors into to maximize the consistency of error features within groups and the differences in error features between groups. The model can calculate the feature aggregation index under different numbers of divisions and ultimately map it to an integer through the output layer; this integer is the K value that optimally represents the current error distribution structure.

[0068] Step S4: Based on the multiple ball-playing error points and the K value, clustering is performed to determine K error regions.

[0069] In some embodiments, Figure 5 This is a flowchart illustrating the process of determining K error regions according to an embodiment of the present invention, wherein steps S41-S42 for determining the K error regions are as follows:

[0070] Step S41: Based on the multiple ball-hitting error points and the K value, use the K-means clustering algorithm to determine K clusters.

[0071] K-Means clustering is an iterative clustering algorithm. It divides a dataset into disjoint subsets by setting a predefined number of clusters, minimizing the sum of squared distances between each data point in a subset and the centroid of that subset. K-Means clustering converges quickly and automatically groups data points with similar spatial characteristics into the same cluster.

[0072] K clusters are specific data groups formed by dividing all ball-playing errors according to feature similarity using the K-means clustering algorithm.

[0073] Each cluster contains a group of multiple hitting errors that are highly similar in spatial location, error type, or action characteristics. Different clusters exhibit significant differences in spatial distribution and error characteristics.

[0074] The process of clustering multiple ball-playing errors using the K-means clustering algorithm and the K-value is as follows: First, the three-dimensional spatial coordinate data of multiple ball-playing errors are extracted as the core clustering basis. Simultaneously, a preset K-value is read to determine the number of target clusters. The algorithm randomly selects K errors as initial cluster centers. Then, it calculates the Euclidean distance from each error to these K cluster centers and assigns each error to the cluster corresponding to its nearest cluster center according to the nearest distance principle. After the first round of assignment, the algorithm recalculates the mean of the three-dimensional spatial coordinates of all errors within each cluster, using this mean as the new cluster center. The distance calculation and error allocation operations are then repeated. This process iterates until the change in the cluster center position between two consecutive calculations is less than a preset threshold. At this point, the clustering process stops, and the final K groups form K clusters.

[0075] Clustering effectively integrates complex data on multiple hitting errors. Because these errors encompass multi-dimensional information such as spatial location, error type, and striking motion characteristics, and are numerous and scattered, direct analysis of this data makes it difficult to discover patterns in athlete errors. Clustering, however, groups errors with similar characteristics into K clusters, significantly simplifying the data structure and facilitating the extraction of valuable error distribution and feature correlation information. Dividing multiple hitting errors into K clusters visually displays the different concentrated areas and characteristic types of athlete errors on the court. Analysis of these clusters quickly identifies the dominant error types and typical motion deviations in each area, and further comparison of the characteristics of different clusters clearly reveals the athlete's technical weaknesses and error causes in different court areas.

[0076] Step S42: Connect the points of hitting errors at the edges of each cluster to obtain K error areas.

[0077] The edge of each cluster refers to the point where a player makes a mistake within a single cluster obtained by the K-means clustering algorithm, and the point is located at the outer boundary of the spatial distribution of that cluster. The three-dimensional spatial coordinates of these points are relatively far from the cluster center.

[0078] The K error regions are closed spatial regions formed by connecting the edge of each cluster with the points of error in play according to their actual spatial positions on the court. Each error region corresponds to a cluster obtained from the clustering, and the number of error regions is consistent with the number of clusters K obtained from the clustering.

[0079] In some embodiments, the three-dimensional spatial coordinates of all hitting errors within each cluster can be extracted first, and the hitting errors that form the edges of the cluster distribution contour can be filtered out. Then, the hitting errors on these edges can be connected in order of their spatial positions on the court to form a closed polygonal region. Each cluster generates a corresponding error region, and finally, K error regions matching the number of clusters are obtained.

[0080] Step S5: Construct a fault region map, which includes K fault region nodes and multiple edges between the K fault region nodes. The node characteristics of each fault region node are multiple ball-playing errors in this fault region, and the edges between fault region nodes are the straight-line distances between fault regions.

[0081] A fault region map is a graphical data structure that represents the spatial distribution structure of faults and the relationships between regions in a badminton match. A fault region map contains information on K fault region nodes describing region attributes, as well as edge information describing the spatial adjacency relationships between regions.

[0082] Each fault region node records in detail all the hitting errors contained within that region. Edges represent the straight-line distances between fault regions, and the edges in the fault region map can intuitively show the positional characteristics of each fault region and their spatial distance relationships.

[0083] Step S6: Process the error region map based on the graph neural network to determine at least K hitting training points, wherein each error region has at least one hitting training point.

[0084] Graph Neural Networks (GNNs) are deep learning architectures specifically designed for processing topologically structured data such as graphs. GNNs enable information exchange and aggregation between nodes through message passing mechanisms. In a GNN, each node can update its state representation based on its own features and those of its neighbors. GNNs can simultaneously capture the topological structure of the graph and the attribute features of the nodes. Through multi-layer graph convolution or aggregation operations, GNNs can deeply mine the dependencies between nodes and global distribution patterns.

[0085] A hitting training point is a specific location within a fault region, determined by a graph neural network after processing the fault region map, for targeted training by the athlete. Each fault region contains at least one hitting training point, and there are at least K hitting training points across K fault regions.

[0086] By constructing a fault region map, the spatial relationship network between K fault regions can be clearly reflected. This relationship information is crucial for determining the hitting training point, as the spatial relationship between different fault regions affects the correlation analysis of fault causes. Using multiple hitting errors in each fault region as node features and the straight-line distance between fault regions as edge features allows for more efficient use of data information. This helps the graph neural network better understand the feature differences and spatial relationships between each fault region, thereby improving the accuracy of hitting training point determination. Processing fault region map data based on graph neural networks can effectively learn the complex relationships and information transmission between nodes, thus more accurately uncovering the core fault features and training priorities of each fault region. Compared to traditional clustering analysis or single feature extraction methods, graph neural networks have better representation and learning capabilities when processing graph data, and can take into account both the individual characteristics of fault regions and the overall relationships between regions.

[0087] Graph neural networks (Graph Neural Networks) can aggregate the features of each node and its neighboring nodes through graph convolutional layers, and analyze the impact of spatial distance between different error regions on the occurrence of errors, such as whether there is a correlation between the causes of errors in adjacent error regions. Graph Neural Networks perform in-depth feature mining on the nodes of each error region to calculate the dense center location of error points within the region. This dense center location can be the area where athletes make the most errors, thus serving as the primary hitting training point. If a certain error region is large or the error points are scattered, the model will determine additional auxiliary training points within that region based on the distribution density of error points and the concentration of error types, ensuring that each error region has at least one hitting training point.

[0088] Step S7: Determine the target training plan based on the at least K ball-hitting training points.

[0089] In some embodiments, Figure 6 This is a flowchart illustrating the process of determining a target training plan according to an embodiment of the present invention, wherein steps S71-S73 are as follows:

[0090] Step S71: Based on the badminton match video of the athlete and the at least K hitting training points, determine multiple high-importance hitting training points and multiple medium-importance hitting training points.

[0091] In some embodiments, a hit training point determination model can be used to determine multiple high-importance hit training points and multiple medium-importance hit training points. The hit training point determination model is a recurrent neural network model. The input to the hit training point determination model is the badminton match video of the athlete and the at least K hit training points, and the output of the hit training point determination model is the multiple high-importance hit training points and the multiple medium-importance hit training points.

[0092] Multiple high-importance hitting training points are identified by analyzing badminton match videos and at least K hitting training points through a hitting training point determination model. These points are those that have a decisive impact on the outcome of the match or have serious consequences for mistakes.

[0093] The types of mistakes made at key hitting points can directly affect the outcome of a match, and these key hitting points are the core problems among an athlete's technical weaknesses.

[0094] Multiple medium-important hitting training points are training points whose importance is less than that of high-important hitting training points, determined by the hitting training point determination model. These points have an impact on the smoothness of the game but a relatively low probability of causing direct point loss.

[0095] Targeted training of key hitting points can help improve athletes' technical stability and solve minor errors in matches.

[0096] The badminton match videos of athletes contain temporal information such as the scenarios in which errors occur, the frequency of errors, and the degree of impact on the game situation corresponding to each hitting training point. The hitting training points clearly define the specific court locations that need to be analyzed. This information can provide a complete data analysis basis for the model to distinguish the importance of training points. The model can determine the impact weight of different training points on athletes by analyzing the match details in the badminton match videos.

[0097] Recurrent Neural Networks (RNNs) can leverage their temporal analysis capabilities to trace back match segments in badminton match videos that are spatially related to at least K practice points. RNNs can analyze whether a player's errors near these practice points occur during crucial scoring phases, whether the errors are caused by unforced variations of movements, and whether the errors directly lead to points for the opponent. Through a memory mechanism, RNNs can associate the match's contextual information and calculate the risk cost of each error at a practice point. If errors associated with a particular practice point frequently occur at match points or lead to consecutive lost points, the model will determine that practice point has extremely high training priority and classify it as one of the multiple high-importance practice points. If errors associated with a particular practice point are mostly forced errors and do not cause serious tactical disadvantages, the model will classify this practice point as one of the multiple medium-importance practice points.

[0098] Step S72: Generate multiple training plans based on the multiple high-importance hitting training points, the multiple medium-importance hitting training points, and the at least K hitting training points.

[0099] In some embodiments, a plan determination model can be used to generate multiple training plans. The plan determination model is a deep neural network model. The inputs to the plan determination model are the plurality of high-importance hitting training points, the plurality of medium-importance hitting training points, and the at least K hitting training points; the output of the plan determination model is the multiple training plans.

[0100] Multiple training plans are various badminton training implementation schemes output by the planning model. Each training plan includes the training frequency of each hitting point, the specific action training content for each hitting point, the training sequence of each hitting point, and the single training duration of each hitting point.

[0101] The training frequency for each hitting practice point refers to the number of training sessions scheduled for a single hitting practice point within a single training plan. The training frequency for high-importance hitting practice points is higher than that for medium-importance hitting practice points.

[0102] The specific training content for each hitting point is designed based on the corresponding technical weaknesses and error types in badminton techniques. The training content includes practice on controlling the racket face angle for net shots, aiming for baseline smashes, adjusting stance for receiving serves, controlling the power of backcourt drops, and timing the power of midcourt lobs.

[0103] The training sequence of each hitting training point refers to the order in which different specific hitting training points are carried out during the training process.

[0104] At least K hitting practice points cover all the court locations that athletes need to train on. Among them, several high-importance hitting practice points clarify the core priority of training, while several medium-importance hitting practice points define the direction of auxiliary training. This data clarifies the key areas and coverage of training and provides a basis for the planning model to formulate scientific training plans. The model can allocate reasonable training resources and generate diverse training programs based on the importance of the training points.

[0105] Deep neural networks can generate multiple training plans due to their multi-dimensional feature extraction and precise resource allocation capabilities. They leverage powerful feature mapping and non-linear combination capabilities to process input spatial location and attribute data. The model first deeply analyzes the detailed feature data behind each input hitting training point, including hitting action characteristics and error types. Through the neural computation of hidden layers, the model establishes a precise mapping relationship between these specific action defect features and preset kinematic correction strategies. For example, when the model identifies a high-importance hitting training point associated with the feature "excessive racket angle leading to a high return," it automatically matches and generates "racket angle control practice for net drop shots" as the specific action training content for that point. Simultaneously, deep neural networks can calculate weight allocation based on the importance labels of input points. The model can adjust the output layer values ​​through activation functions, based on the principle that the weights of multiple high-importance hitting training points are significantly higher than those of multiple medium-importance hitting training points, thereby allocating higher training frequency and longer single training durations to high-importance points. When determining the training sequence for each hitting practice point, the model comprehensively considers the spatial distance of the hitting practice points on the court and the rationality of physical exertion. By simulating the athlete's movement path, it generates a sequence that conforms to ergonomics and can simulate the rhythm of actual combat. In addition, by adjusting internal hyperparameters or adopting a multi-output structure, the model can simulate different training orientations. For example, the model can generate a skill-oriented plan that focuses on correcting specific technical movements, another physical fitness plan that focuses on high-frequency movement and positioning throughout the court, and a comprehensive plan that focuses on alternating practice of high-importance and medium-importance hitting practice points, ultimately outputting multiple differentiated and feasible training plans.

[0106] Step S73: Display the multiple training plans on the screen, obtain the training plan selected by the user, and use the training plan selected by the user as the target training plan.

[0107] A target training plan is a scheme selected by the user from multiple training plans for the actual training of athletes. The target training plan can be tailored to the athlete's error characteristics and training needs.

[0108] Based on the same inventive concept Figure 7 A schematic diagram of a sports event training prediction system based on AI and image analysis is provided as an embodiment of the present invention. The sports event training prediction system based on AI and image analysis includes:

[0109] Module 81 is used to acquire badminton match videos of athletes;

[0110] The error point determination module 82 is used to determine multiple badminton error points based on the badminton match video of the athlete using the error point determination model.

[0111] Error point analysis module 83 is used to determine the K value based on the multiple hitting error points;

[0112] Clustering module 84 is used to cluster and determine K error regions based on the multiple ball-playing error points and the K value;

[0113] The graph construction module 85 is used to construct a fault region graph, which includes K fault region nodes and multiple edges between the K fault region nodes. The node characteristics of each fault region node are multiple ball-playing mistakes in this fault region, and the edges between the fault region nodes are the straight-line distance between the fault regions.

[0114] The training point determination module 86 is used to process the error region map based on the graph neural network to determine at least K hitting training points, wherein each error region has at least one hitting training point;

[0115] Training plan determination module 87 is used to determine a target training plan based on the at least K hitting training points.

[0116] It should be noted that, in order to simplify the descriptions disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments of this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0117] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A method for predicting sports event training based on AI and image analysis, characterized in that, include: Obtain videos of athletes playing badminton; Based on the badminton match video of the athlete, a fault point determination model is used to identify multiple hitting fault points. Each fault point corresponds to a three-dimensional spatial coordinate at the time of the fault and includes the timestamp of the fault, the fault type, the characteristics of the hitting action corresponding to the fault, the athlete's standing area at the time of the fault, and the trajectory parameters of the shuttlecock at the time of the fault. The fault point determination model is a recurrent neural network model. The process of determining multiple hitting fault points based on the badminton match video of the athlete includes: Based on the badminton match video of the athlete, determine the time sequence parameter set of multiple single shuttlecock return, the set of real-time flight trajectory sequence of the badminton shuttlecock for multiple single shuttlecock return, and the video frame corresponding to the abnormal action; Based on the set of athlete action timing parameters for multiple single-shot returns, the set of badminton shuttlecock real-time flight trajectory sequences for multiple single-shot returns, the time domain value of athlete's center of gravity adjustment frequency determined by the video frames corresponding to the abnormal actions, the athlete's swing angular velocity change curve, the athlete's movement start speed time history data, and the single-shot action scoring table. Based on the set of real-time flight trajectory sequences of badminton shuttlecocks from multiple single-shot returns, the time domain value of the athlete's center of gravity adjustment frequency, the change curve of the athlete's swing angular velocity, the time history data of the athlete's movement start speed, and the single-shot hitting action scoring table, multiple hitting error points are determined. The K value is determined based on the multiple points of hitting error; Based on the multiple points of hitting errors and the K value, clustering is performed to determine K error regions; Construct a fault region map, which includes K fault region nodes and multiple edges between the K fault region nodes. The node characteristics of each fault region node are multiple hitting errors in this fault region, and the edges between the fault region nodes are the straight-line distances between the fault regions. The error region map is processed using a graph neural network to determine at least K hitting training points, wherein each error region has at least one hitting training point. A target training plan is determined based on the at least K ball-hitting training points.

2. The sports event training prediction method based on AI and image analysis as described in claim 1, characterized in that, The determination of the target training plan based on the at least K hitting training points includes: Based on the badminton match video of the athlete and the at least K hitting training points, multiple high-importance hitting training points and multiple medium-importance hitting training points are determined. Multiple training plans are generated based on the multiple high-importance hitting training points, the multiple medium-importance hitting training points, and the at least K hitting training points; The multiple training plans are displayed on the screen, and the training plan selected by the user is obtained and used as the target training plan.

3. The sports event training prediction method based on AI and image analysis as described in claim 1, characterized in that, The step of clustering and determining K error regions based on the multiple hitting errors and the K value includes: Based on the multiple points of hitting errors and the K value, the K-means clustering algorithm is used to determine K clusters; Connect the points of hitting errors at the edges of each cluster to obtain K error zones.

4. A sports event training prediction system based on AI and image analysis, characterized in that, include: The acquisition module is used to acquire videos of athletes playing badminton matches. The error point determination module is used to determine multiple hitting errors based on the badminton match video of the athlete, using an error point determination model. Each hitting error point corresponds to a three-dimensional spatial coordinate at the time of the error and includes a timestamp of the error, error type, the corresponding hitting action characteristics, the athlete's standing area at the time of the error, and the trajectory parameters of the shuttlecock at the time of the error. The error point determination model is a recurrent neural network model. Determining multiple hitting errors based on the badminton match video of the athlete includes: Based on the badminton match video of the athlete, determine the time sequence parameter set of multiple single shuttlecock return, the set of real-time flight trajectory sequence of the badminton shuttlecock for multiple single shuttlecock return, and the video frame corresponding to the abnormal action; Based on the set of athlete action timing parameters for multiple single-shot returns, the set of badminton shuttlecock real-time flight trajectory sequences for multiple single-shot returns, the time domain value of athlete's center of gravity adjustment frequency determined by the video frames corresponding to the abnormal actions, the athlete's swing angular velocity change curve, the athlete's movement start speed time history data, and the single-shot action scoring table. Based on the set of real-time flight trajectory sequences of badminton shuttlecocks from multiple single-shot returns, the time domain value of the athlete's center of gravity adjustment frequency, the change curve of the athlete's swing angular velocity, the time history data of the athlete's movement start speed, and the single-shot hitting action scoring table, multiple hitting error points are determined. The error point analysis module is used to determine the K value based on the multiple hitting error points; The clustering module is used to cluster and determine K error regions based on the multiple hitting error points and the K value; The graph construction module is used to construct a fault region graph, which includes K fault region nodes and multiple edges between the K fault region nodes. The node characteristics of each fault region node are multiple hitting errors in this fault region, and the edges between the fault region nodes are the straight-line distance between the fault regions. The training point determination module is used to process the error region map based on a graph neural network to determine at least K hitting training points, wherein each error region has at least one hitting training point; The training plan determination module is used to determine a target training plan based on the at least K hitting training points.

5. The sports event training prediction system based on AI and image analysis as described in claim 4, characterized in that, The training plan determination module is also used for: Based on the badminton match video of the athlete and the at least K hitting training points, multiple high-importance hitting training points and multiple medium-importance hitting training points are determined. Multiple training plans are generated based on the multiple high-importance hitting training points, the multiple medium-importance hitting training points, and the at least K hitting training points; The multiple training plans are displayed on the screen, and the training plan selected by the user is obtained and used as the target training plan.

6. The sports event training prediction system based on AI and image analysis as described in claim 4, characterized in that, The clustering module is also used for: Based on the multiple points of hitting errors and the K value, the K-means clustering algorithm is used to determine K clusters; Connect the points of hitting errors at the edges of each cluster to obtain K error zones.

Images