Football training action evaluation and optimization method and system fusing multi-modal data

Abstract

The present application relates to the technical field of intelligent football training, solves the problem that the multi-source training data cannot be effectively processed in the football training process in the prior art, the football training action evaluation accuracy is low, and provides a football training action evaluation and optimization method and system fusing multi-modal data. The method comprises the following steps: acquiring training video data, motion data, trainer identity data and environment data; the training video data, the motion data, the trainer identity data and the environment data are time-synchronized and aligned, and are fused and processed to obtain fused data; posture features, time sequence features and motion dynamic features are extracted respectively, and feature-level fusion is performed to obtain fused features; the fused features are input into a football training action evaluation model, and based on the action evaluation result, an action optimization suggestion is generated. The present application solves the problems that the existing football training action evaluation is highly subjective, the data is not fully utilized, and effective optimization guidance cannot be provided.

Description

Technical Field

[0001] This invention relates to the field of intelligent football training technology, and in particular to a method and system for evaluating and optimizing football training movements by integrating multimodal data. Background Technology

[0002] As football becomes increasingly professional and scientific in its training, trainees have a growing need for quantitative assessments of movement standardization, execution quality, and training effectiveness during daily training. Objective and detailed evaluation of football training movements, coupled with targeted suggestions for optimization, helps trainees identify deviations and improve training methods, thereby increasing training efficiency and reducing the risk of sports injuries. Therefore, accurate evaluation and optimization of football training movements is of significant practical importance.

[0003] Existing techniques for evaluating and optimizing football training movements primarily rely on coach experience and observation, single video playback analysis, or simple motion parameter statistics based on wearable sensors. On the one hand, evaluation methods based on manual observation or video playback are highly subjective and struggle to quantify complex, continuous movements. On the other hand, evaluation methods based on single sensor data typically focus only on exercise intensity or a single dynamic indicator, lacking a comprehensive analysis of spatial posture changes, temporal continuity, and the influence of environmental factors. Furthermore, existing technologies often lack a unified time alignment and fusion mechanism for data from different sources, resulting in insufficient correlation between multimodal data. This, in turn, affects the accuracy and stability of movement evaluation results, making it difficult to generate targeted movement optimization suggestions.

[0004] Therefore, how to effectively process multi-source training data during football training to achieve accurate evaluation and optimization guidance of football training movements is a key technical problem that urgently needs to be solved. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for evaluating and optimizing football training movements by integrating multimodal data, in order to solve the problem that the existing technology cannot effectively process multi-source training data during football training and has low accuracy in evaluating football training movements.

[0006] The technical solution adopted in this invention is: In a first aspect, the present invention provides a method for evaluating and optimizing football training movements by integrating multimodal data, the method comprising: Acquire training video data, motion data, trainee identity data, and environmental data of the target trainee during football training, wherein the motion data includes motion parameter data collected by wearable sensors; Based on the timestamp information of the training video data and the motion data, the training video data, the motion data, the trainee identity data and the environmental data are time-synchronized and aligned, and then fused after time alignment to obtain fused data corresponding to the football training process. The posture features used to characterize the spatial posture changes of football training movements, the temporal features used to characterize the continuous change law of football training movements, and the motion dynamic features used to characterize the intensity and stability of football training movements are extracted from the fused data respectively. The posture features, the temporal features, and the motion dynamic features are then fused at the feature level to obtain the fused features used to characterize the characteristics of football training movements. The fused features are input into a pre-trained football training action evaluation model based on a deep learning architecture for inference analysis, and the action evaluation results corresponding to the football training actions are output. The action evaluation results include action standardization score and action completion quality score. Based on the action evaluation results, action optimization suggestions are generated corresponding to the football training actions. These suggestions are used to guide the target trainee to adjust their football training actions.

[0007] In an optional embodiment, acquiring training video data, movement data, trainee identity data, and environmental data of the target trainee during football training includes: The training movements of the target trainee during football training are filmed to obtain a series of multiple video images that record the football training process of the target trainee. The video images are then processed by frame serialization to obtain the training video data corresponding to the training. Wearable inertial sensors are fixedly worn on the lower limbs of the target trainee. During football training, the three-axis acceleration data and three-axis angular velocity data of the target trainee are collected at a preset sampling frequency. The collected data are formatted to obtain the corresponding motion parameter data as the motion data. Based on the unique identification information corresponding to the target trainee, the trainee identity data corresponding to the target trainee is read from the pre-established trainee information database; By installing an environmental monitoring module at the football training field, environmental parameters during football training are collected to obtain the environmental data. The environmental parameters include at least ambient temperature and ambient humidity.

[0008] In an optional embodiment, the step of performing time synchronization alignment on the training video data, the movement data, the trainee identity data, and the environmental data based on the timestamp information of the training video data and the movement data, and then performing fusion processing after time alignment to obtain fused data corresponding to the football training process includes: Extract the first timestamp information corresponding to each frame of video image in the training video data, and extract the second timestamp information corresponding to each sampling point in the motion data; Based on the first timestamp information and the second timestamp information, a unified training timeline for characterizing the football training process is constructed. According to the unified training timeline, the training video data and the motion data are subjected to time mapping processing so that each frame of video image and the corresponding motion data at the time dimension are established. When the sampling frequencies of the training video data and the motion data are inconsistent on the unified training time axis, interpolation or resampling is performed on the data with the lower sampling frequency to achieve synchronous alignment of the training video data and the motion data in the time dimension. The training video data and motion data that have been synchronized and aligned in time are associated and fused with the trainee identity data and environmental data within the corresponding training time period to obtain fused data corresponding to the football training process.

[0009] In an optional embodiment, the step of associating and fusing the training video data and motion data that have completed time synchronization with the trainee identity data and environmental data within the corresponding training time period to obtain fused data corresponding to the football training process includes: Based on the unified training timeline, the target training time period corresponding to the training video data and the motion data that have completed time synchronization alignment is determined; The target training time period is matched with the trainee identity data to determine the target trainee identity information corresponding to the target training time period; Based on the target training time period, filter environmental state data corresponding to the target training time period from the environmental data; Using the unified training timeline as a time index, the training video data, the motion data, the target trainee identity information, and the environmental state data are organized to construct a multimodal data structure corresponding to the football training process. The multimodal data structure is fused to generate fused data corresponding to the football training process.

[0010] In an optional embodiment, the step of extracting posture features for characterizing spatial posture changes in football training movements, temporal features for characterizing continuous change patterns in football training movements, and motion dynamic features for characterizing the intensity and stability of football training movements from the fused data, and then performing feature-level fusion of the posture features, temporal features, and motion dynamic features to obtain fused features for characterizing the characteristics of football training movements includes: Based on the training video data in the fused data, key body parts of the target trainee during football training are detected to obtain the spatial position information of each key body part during continuous training. Combined with the body parameter information in the target trainee's identity information, the spatial position information is normalized to obtain the posture features. Based on the training video data and motion data in the fused data, the movement change trend of the target trainee's football training movements during continuous training is analyzed, and the movement change trend is corrected by combining the environmental state data to extract the temporal features. Based on the motion data in the fused data, the amplitude and fluctuation characteristics of the motion parameters of the target trainee during football training are analyzed, and the changes in the motion parameters are compensated by combining the environmental state data to extract the motion dynamic features. The posture features, temporal features, and motion dynamic features are subjected to feature scale unification processing. After the feature scale unification processing is completed, the posture features, temporal features, and motion dynamic features are fused to generate fused features that characterize the characteristics of football training movements.

[0011] In an optional embodiment, based on the training video data in the fused data, key body parts of the target trainee during football training are detected to obtain spatial position information of each key body part at continuous training moments. This spatial position information is then normalized by combining it with body parameter information from the target trainee's identity information to obtain the posture features, which include: The video images in each frame of the training video data are preprocessed, and the preprocessing includes at least one image denoising, image cropping or image scaling; Based on the preprocessed video images of each frame, the key body parts of the target trainee are identified and located, and the two-dimensional or three-dimensional coordinate information of each key body part in the corresponding video frame is determined. According to the time sequence of video frames, the coordinate information of the key body parts is arranged in time sequence to obtain the spatial position information sequence of each key body part at continuous training time. Obtain body parameter information related to body structure from the target trainee's identity information, wherein the body parameter information includes at least height, weight or limb length parameters; Based on the body parameter information, the spatial position information sequence is normalized to obtain posture features that characterize the spatial posture changes of football training movements.

[0012] In an optional embodiment, the step of analyzing the movement change trend of the target trainee's football training movements during continuous training based on the training video data and motion data in the fused data, and correcting the movement change trend by combining the environmental state data, and extracting the temporal features includes: From the fused data, training video data and motion data corresponding to the target trainee at continuous training moments are extracted to construct a motion time sequence data sequence for characterizing the football training process. Based on the action time sequence data, action change parameters between adjacent training moments are extracted. The action change parameters include at least one of the following: joint angle change, limb displacement change, action execution rhythm change, and movement speed change. Based on the motion change parameters, analyze the change trend of the target trainee's football training motion under the continuous training time to obtain the initial motion temporal characteristics; Based on the environmental state data, environmental impact correction processing is performed on the initial action timing characteristics to determine the timing characteristics; The process involves analyzing the amplitude and fluctuation characteristics of movement parameters of the target trainee during football training based on the motion data in the fused data, and compensating for the changes in the movement parameters by combining the environmental state data, thereby extracting the dynamic features of the movement, including: Based on the motion data, the variation amplitude and fluctuation characteristics of each motion parameter of the target trainee during continuous training are calculated to obtain the initial motion dynamic parameters; Based on the initial motion dynamic parameters, analyze the changes in motion intensity and motion stability characteristics of the target trainee's football training movements during the training process; Based on the environmental state data, environmental compensation processing is performed on the initial motion dynamic parameters to determine the motion dynamic features.

[0013] In an optional embodiment, the step of performing feature scale unification processing on the posture features, the temporal features, and the motion dynamic features, and then fusing the posture features, the temporal features, and the motion dynamic features after completing the feature scale unification processing to generate fused features for characterizing the characteristics of football training movements includes: The posture features, temporal features, and motion dynamic features are subjected to feature scale unification processing to obtain scale-unified posture features, temporal features, and motion dynamic features; Based on the target trainee's identity information and the environmental status data, the feature weight parameters of various features in the evaluation of football training actions are determined. Based on the feature weight parameters, the scale-unified pose features, temporal features, and motion dynamic features are weighted and fused to generate initial fused features; The initial fused features are processed by feature concatenation, dimensionality reduction, or feature selection to obtain the fused features.

[0014] In an optional embodiment, the fused features are input into a pre-trained deep learning-based football training action evaluation model for inference analysis, and the model outputs action evaluation results corresponding to the football training actions. The action evaluation results include action standardization scores and action completion quality scores, including: The fused features are processed to conform to the input dimension and data structure requirements of the football training action evaluation model. The fused features are then input into the pre-trained football training action evaluation model based on a deep learning architecture, and forward inference calculations are performed to obtain intermediate action evaluation feature representations corresponding to the football training actions. Based on the intermediate action evaluation feature representation, a normative evaluation branch for characterizing the standardization of football training actions and a quality evaluation branch for characterizing the completion quality of football training actions are constructed respectively. The intermediate action evaluation feature representation is analyzed through the normative evaluation branch, and the action normative score corresponding to the football training action is output. The intermediate action evaluation feature representation is analyzed through the quality evaluation branch, and the action completion quality score corresponding to the football training action is output. The action standardization score and the action completion quality score are combined to form the action evaluation result of the football training action.

[0015] Secondly, embodiments of the present invention also provide a football training action evaluation and optimization system that integrates multimodal data, including: at least one processor, at least one memory, and computer program instructions stored in the memory, which implement the method of the first aspect described above when the computer program instructions are executed by the processor.

[0016] In summary, the beneficial effects of the present invention are as follows: The present invention provides a method and system for evaluating and optimizing football training movements by fusing multimodal data. The method includes: acquiring training video data, motion data, trainee identity data, and environmental data of a target trainee during football training, wherein the motion data includes motion parameter data collected by wearable sensors; performing time synchronization alignment on the training video data, motion data, trainee identity data, and environmental data based on the timestamp information of the training video data and the motion data, and performing fusion processing after time alignment to obtain fused data corresponding to the football training process; and extracting posture features for characterizing spatial posture changes in football training movements and other features for characterizing spatial posture changes in football training movements from the fused data. This invention constructs a multimodal data foundation covering movement performance, individual differences, and external conditions by simultaneously acquiring training video data, motion data collected by wearable sensors, and trainee identity and environmental data related to the training process. Furthermore, it utilizes timestamp information from training video data and motion data to perform synchronous alignment and fusion processing of various data types under a unified timeline, effectively solving the problems of inconsistent time and insufficient correlation between multiple data sources in existing technologies. Building upon this foundation, multidimensional features reflecting spatial posture changes, temporal continuity, and intensity and stability of movements are extracted from the fused data. These features are then fused at the feature level to form a holistic representation of football training movements, thus avoiding the bias inherent in evaluations based on single data or metrics. Subsequently, the fused features are input into a pre-trained deep learning movement evaluation model for inference analysis. This model outputs objective and quantitative scores for movement standardization and completion quality, taking into account multimodal information, significantly improving the accuracy and consistency of the evaluation results. Finally, targeted movement optimization suggestions are generated based on the evaluation results, enabling trainees to adjust their training movements with a clear basis. This achieves a closed-loop process from data collection and evaluation analysis to movement optimization, effectively addressing the technical problems of strong subjectivity, insufficient data utilization, and difficulty in providing effective optimization guidance in existing football training movement evaluations. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, and these are all within the protection scope of the present invention.

[0018] Figure 1 This is a schematic diagram illustrating the overall workflow of the football training motion evaluation and optimization method that integrates multimodal data in Embodiment 1 of the present invention; Figure 2 This is a flowchart illustrating the process of extracting posture features, temporal features, and motion dynamic features in Embodiment 1 of the present invention, and performing feature-level fusion to obtain fused features. Figure 3 This is a schematic diagram of the structure of the football training action evaluation and optimization system that integrates multimodal data in Embodiment 2 of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. In the description of the present invention, it should be understood that the terms "center," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Where there is no conflict, embodiments of the present invention and the various features thereof can be combined with each other, all of which are within the scope of protection of the present invention. Example 1

[0020] Please see Figure 1Embodiment 1 of the present invention discloses a method for evaluating and optimizing football training movements by fusing multimodal data, the method comprising: Acquire training video data, motion data, trainee identity data, and environmental data of the target trainee during football training, wherein the motion data includes motion parameter data collected by wearable sensors; Specifically, by collecting multi-source information from the actual football training process of the target trainees, foundational data support is provided for subsequent movement evaluation. Training video data is used to fully record the trainees' movement performance during training, typically acquired in the form of continuous multi-frame images or video streams, reflecting the spatial posture and detailed changes in movement. Motion data is collected by wearable sensors worn on pre-designed parts of the trainees' bodies; this type of motion parameter data generally includes parameters reflecting body motion states, such as acceleration and angular velocity, used to describe the dynamic characteristics of training movements. Trainee identity data is used to characterize the differences in physical conditions and training backgrounds among different trainees, providing individualized references for subsequent movement analysis. Environmental data reflects the objective external conditions during training, such as ambient temperature and humidity. By simultaneously acquiring these multiple types of data during the same training process, the collected data not only covers the movements themselves but also incorporates individual characteristics and environmental factors, laying a comprehensive and reliable data foundation for subsequent multimodal data fusion and movement evaluation.

[0021] Based on the timestamp information of the training video data and the motion data, the training video data, the motion data, the trainee identity data and the environmental data are time-synchronized and aligned, and then fused after time alignment to obtain fused data corresponding to the football training process. Specifically, since training video data and motion data are typically acquired by different devices, their sampling frequencies and time bases differ. Therefore, this step utilizes the timestamp information inherent in various data sources to perform synchronization alignment processing on a unified timeline for different data sources. In practice, by extracting the time information corresponding to each frame in the training video and the time information of each sampling point in the motion data, a unified training time series is established, enabling a clear temporal correspondence between video frames at the same training moment and their corresponding motion parameters. After time synchronization is completed, the time-aligned training video data and motion data are associated and integrated with trainee identity data and environmental data that remain relatively stable within the same training time period, forming structured fused data. This processing method effectively avoids the problem of time misalignment or fragmented use of multi-source data in existing technologies, allowing different modalities of data to support and complement each other in the same time dimension, thereby improving the accuracy of subsequent feature extraction and motion analysis.

[0022] The posture features used to characterize the spatial posture changes of football training movements, the temporal features used to characterize the continuous change law of football training movements, and the motion dynamic features used to characterize the intensity and stability of football training movements are extracted from the fused data respectively. The posture features, the temporal features, and the motion dynamic features are then fused at the feature level to obtain the fused features used to characterize the characteristics of football training movements. Specifically, after obtaining the fused data, this step focuses on feature extraction processing based on the multidimensional performance characteristics of football training movements. First, by analyzing information related to training videos in the fused data, features reflecting spatial changes in body posture during training movements are extracted to describe the relative positional relationships and changes of various body parts during the movement. Second, combining the temporal continuity of video and motion data, the changing patterns of training movements at continuous training moments are analyzed to extract temporal features reflecting the rhythm, coherence, and sequence of movements. Simultaneously, by analyzing the amplitude and stability characteristics of various motion parameters in the motion data, dynamic features reflecting movement intensity and control stability are extracted. Subsequently, feature-level fusion processing is performed on these multiple features, enabling movement information from different dimensions to form a unified expression within the same feature space, thereby obtaining fused features that comprehensively characterize the characteristics of football training movements. This processing method effectively overcomes the information gaps caused by single-feature or single-modal evaluations, providing more complete and reliable input features for subsequent deep learning-based movement evaluation.

[0023] The fused features are input into a pre-trained football training action evaluation model based on a deep learning architecture for inference analysis, and the action evaluation results corresponding to the football training actions are output. The action evaluation results include action standardization score and action completion quality score. Specifically, the aforementioned fused features are used as input to a pre-trained football training motion evaluation model to analyze the training motions. This pre-trained model is built on a deep learning architecture and learns the mapping relationship between fused features and motion quality by training on a large number of labeled football training motion samples. During actual inference, the fused features are input to the model's feature processing and discrimination layers. The model comprehensively analyzes the spatial posture information, temporal variation patterns, and dynamic characteristics contained in the features, thereby outputting a motion evaluation result corresponding to the current training motion. The motion evaluation result includes a motion standardization score and a motion completion quality score. The motion standardization score reflects the degree of matching between the training motion and the standard motion pattern, while the motion completion quality score measures the performance of the training motion in terms of stability, coherence, and overall execution effect. By adopting a deep learning-based evaluation method, the subjective influence of human evaluation can be reduced, making the motion evaluation results more objective, stable, and possessing good generalization ability.

[0024] Based on the action evaluation results, action optimization suggestions are generated corresponding to the football training actions. These suggestions are used to guide the target trainee to adjust their football training actions.

[0025] Specifically, after obtaining the movement evaluation results, this step further utilizes these results to conduct targeted analysis of the training movements and generate optimization suggestions to guide trainees in adjusting their movements. In practice, based on the performance of movement standardization and movement completion quality scores across different evaluation dimensions, deviations or weaknesses in the training movements are identified. Combined with pre-set movement optimization strategies, adjustment directions or improvement suggestions are proposed for key aspects of the training movements. The generated movement optimization suggestions directly address specific problems in the training movements, such as deficiencies in range of motion, rhythm, or stability, enabling trainees to clearly identify areas for improvement. By combining the movement evaluation results with the optimization suggestion generation process, a closed-loop processing from data analysis to training guidance is formed. This not only enhances the practical application value of the movement evaluation results but also helps trainees continuously optimize their movement performance during training, thereby improving overall training effectiveness.

[0026] In an optional embodiment, acquiring training video data, movement data, trainee identity data, and environmental data of the target trainee during football training includes: The training movements of the target trainee during football training are filmed to obtain a series of multiple video images that record the football training process of the target trainee. The video images are then processed by frame serialization to obtain the training video data corresponding to the training. Specifically, video image data that fully reflects the changes in training movements is obtained by filming the continuous training movements of the target trainee during football training. The training videos are typically captured by cameras fixed to the training field or by mobile cameras. The shooting angle and frame rate are set according to the characteristics of the training movements to ensure clear recording of key movement phases. After acquiring the video, each frame is processed into a frame sequence, that is, the continuous video is broken down into a series of image frames with a clear temporal order, thus forming a structured video data format. This processing method allows the changes in training movements over time to be accurately expressed in the form of discrete frame sequences, facilitating subsequent time alignment and joint analysis with motion data, and providing a reliable data foundation for the extraction of posture and temporal features.

[0027] Wearable inertial sensors are fixedly worn on the lower limbs of the target trainee. During football training, the three-axis acceleration data and three-axis angular velocity data of the target trainee are collected at a preset sampling frequency. The collected data are formatted to obtain the corresponding motion parameter data as the motion data. Specifically, wearable inertial sensors are fixedly attached to the lower limbs of the target trainee, enabling the sensors to directly sense changes in the lower limb's motion state during training movements. During soccer training, the inertial sensors continuously collect triaxial acceleration and triaxial angular velocity data from the target trainee at a preset sampling frequency to reflect acceleration changes, rotation, and overall motion intensity during training movements. To ensure the uniformity and stability of subsequent processing, the collected raw motion data is formatted, such as by time stamping, standardizing data structure, or performing preliminary outlier processing, thereby forming motion parameter data that can be directly used for time alignment and feature extraction. Motion data acquired in this way can supplement the motion details that video data cannot accurately reflect from a dynamics perspective, helping to improve the comprehensiveness of motion evaluation.

[0028] Based on the unique identification information corresponding to the target trainee, the trainee identity data corresponding to the target trainee is read from the pre-established trainee information database; Specifically, based on the unique identifier of the target trainee, the system retrieves matching identity data from a pre-established trainee information database. This trainee identity data typically includes information related to the trainee's physical condition or training background, used to distinguish individual differences between trainees. Introducing trainee identity data during motion analysis provides a reference for subsequent feature normalization and motion evaluation, making the evaluation results more comparable across different trainees. Centralized management of trainee identity information through a database also facilitates continuous tracking and analysis of the same trainee across multiple training sessions, thereby improving the stability and consistency of the overall evaluation process.

[0029] By installing an environmental monitoring module at the football training field, environmental parameters during football training are collected to obtain the environmental data. The environmental parameters include at least ambient temperature and ambient humidity.

[0030] Specifically, an environmental monitoring module installed at the football training field collects environmental parameters in real time or periodically during football training. These parameters include at least ambient temperature and humidity, reflecting the external conditions during training. Environmental factors can influence a trainee's athletic performance and movement stability to some extent; therefore, including environmental data in the training process data collection helps to make reasonable corrections or compensations for movement performance in subsequent analysis. The environmental data obtained through the monitoring module can be correlated with training video data and motion data over time, providing auxiliary information for subsequent multimodal data fusion, thus making the movement evaluation results more consistent with the actual training scenario.

[0031] In an optional embodiment, the step of performing time synchronization alignment on the training video data, the movement data, the trainee identity data, and the environmental data based on the timestamp information of the training video data and the movement data, and then performing fusion processing after time alignment to obtain fused data corresponding to the football training process includes: Extract the first timestamp information corresponding to each frame of video image in the training video data, and extract the second timestamp information corresponding to each sampling point in the motion data; Specifically, timestamp information reflecting the data acquisition time is extracted from both training video data and motion data. The first timestamp in the training video data typically corresponds to the acquisition time of each frame of the video image, while the second timestamp in the motion data corresponds to the time stamp of motion parameters acquired by the inertial sensor at different sampling points. Since the two types of data come from different sources, their timestamp accuracy and generation methods may differ. Therefore, by extracting and defining their respective timestamp information, a necessary prerequisite is provided for subsequently constructing a unified time reference. This processing transforms video images and motion parameters from isolated data fragments into data units with clear temporal attributes, laying the foundation for establishing temporal correlations between multimodal data.

[0032] Based on the first timestamp information and the second timestamp information, a unified training timeline for characterizing the football training process is constructed. Specifically, after acquiring timestamp information from different data sources, this step constructs a unified training timeline to characterize the entire football training process by comprehensively analyzing the first and second timestamps. This unified training timeline covers the entire training period and can simultaneously accommodate video frame timestamps and motion data sampling timestamps, allowing different modal data to be expressed within the same time coordinate system. This approach eliminates time offset issues caused by factors such as device startup time and differences in sampling mechanisms, ensuring that subsequent processing is based on a unified time benchmark, which is beneficial for improving the accuracy and consistency of multimodal data association.

[0033] According to the unified training timeline, the training video data and the motion data are subjected to time mapping processing so that each frame of video image and the corresponding motion data at the time dimension are established. Specifically, after establishing a unified training timeline, this step performs time mapping processing on the training video data and motion data according to this timeline. This ensures that each frame of the video image corresponds to a specific moment on the unified timeline, while simultaneously mapping each sampling point in the motion data to the same time dimension. Through time mapping processing, the video frame and motion parameters corresponding to a specific training moment can be determined on the unified timeline, thus establishing a clear correspondence in the time dimension. This processing enables subsequent analysis to simultaneously utilize video and motion data to describe the same action phase, effectively avoiding interference caused by time mismatches in action analysis.

[0034] When the sampling frequencies of the training video data and the motion data are inconsistent on the unified training time axis, interpolation or resampling is performed on the data with the lower sampling frequency to achieve synchronous alignment of the training video data and the motion data in the time dimension. Specifically, because training video data and motion data typically have different sampling frequencies during acquisition (e.g., video frame rate is low while motion sensor sampling frequency is high), some time points may only contain one type of data after time mapping. To address this, this step interpolates or resamples the data with lower sampling frequencies to ensure consistency with high-frequency data on a unified training timeline. This process allows for a complete combination of video and motion parameters at the same temporal resolution, avoiding feature loss due to data sparsity and ensuring the continuity and integrity of subsequent fused data in the temporal dimension.

[0035] The training video data and motion data that have been synchronized and aligned in time are associated and fused with the trainee identity data and environmental data within the corresponding training time period to obtain fused data corresponding to the football training process.

[0036] Specifically, after synchronizing and aligning the training video data and motion data in time, this step further integrates and fuses them with trainee identity data and environmental data within the corresponding training time period. Since trainee identity data and environmental data change relatively little within a training cycle, they can be correlated as a whole with the corresponding time-synchronized data according to the training time period, thus forming fused data containing video information, motion parameters, individual characteristics, and environmental conditions. This fusion method allows data from different sources to be uniformly represented within the same data structure, not only improving the sufficiency of data utilization but also providing complete and context-consistent data input for subsequent multi-dimensional feature extraction and motion evaluation, thereby enhancing the overall reliability of motion analysis results.

[0037] In an optional embodiment, the step of associating and fusing the training video data and motion data that have completed time synchronization with the trainee identity data and environmental data within the corresponding training time period to obtain fused data corresponding to the football training process includes: Based on the unified training timeline, the target training time period corresponding to the training video data and the motion data that have completed time synchronization alignment is determined; Specifically, based on the aforementioned unified training timeline, the time range of the training video data and motion data that have already undergone time synchronization and alignment is determined, thereby identifying their start and end times on the unified timeline and defining the corresponding target training period. This method clearly defines which segment of the football training process the data currently used for analysis covers, providing clear time boundaries for subsequent data matching and fusion operations. This process avoids introducing data from irrelevant training phases into the fusion process, helping to improve the accuracy of the correspondence between the fused data and actual training movements.

[0038] The target training time period is matched with the trainee identity data to determine the target trainee identity information corresponding to the target training time period; Specifically, after determining the target training time period, this step matches this time period with trainee identity data. Since trainee identity data typically corresponds one-to-one with trainee identification information, in practice, the target trainees participating in training within that time period can be identified based on the training records or identification information corresponding to that period, and the corresponding target trainee identity information can be extracted accordingly. This matching method ensures that the identity data introduced during the fusion process remains consistent with the actual trainees performing the training actions, thus providing a reliable individualized reference for subsequent feature normalization and action evaluation.

[0039] Based on the target training time period, filter environmental state data corresponding to the target training time period from the environmental data; Specifically, based on the determined target training time period, environmental state data corresponding to that time period is filtered from the environmental data. Environmental data is typically collected continuously in chronological order; therefore, it can be filtered using timestamps or recording times to ensure that the retained environmental state data remains consistent with the time interval in which the training movements occurred. This processing method enables the subsequently fused data to accurately reflect the real environmental conditions during the training movements, providing necessary environmental background information for subsequent analysis of movement trends and dynamic characteristics.

[0040] Using the unified training timeline as a time index, the training video data, the motion data, the target trainee identity information, and the environmental state data are organized to construct a multimodal data structure corresponding to the football training process. Specifically, after filtering identity and environmental data, this step uses a unified training timeline as a time index to perform unified data organization and processing on training video data, motion data, target trainee identity information, and environmental state data. In practice, data from different modalities are arranged according to corresponding moments on the timeline, allowing multiple types of data at the same point in time or within the same time period to form relationships within the data structure. By constructing this multimodal data structure, data from different sources can be managed and accessed within a unified framework, providing a clear data organization form for subsequent fusion processing and improving the overall data processing efficiency and consistency of the system.

[0041] The multimodal data structure is fused to generate fused data corresponding to the football training process.

[0042] Specifically, the aforementioned multimodal data structure undergoes fusion processing, integrating time-aligned video information, motion parameters, target trainee identity information, and environmental state data into a unified fused data format. During the fusion process, data from different modalities can be correlated, integrated, or compressed, allowing them to collectively describe the overall state of football training movements within the same data entity. Through this fusion processing, the resulting fused data can simultaneously contain information on movement performance, motion state, individual characteristics, and environmental conditions, providing complete and context-consistent data input for subsequent feature extraction and movement evaluation, thereby improving the reliability of overall movement analysis and evaluation.

[0043] In an optional embodiment, the step of extracting posture features for characterizing spatial posture changes in football training movements, temporal features for characterizing continuous change patterns in football training movements, and motion dynamic features for characterizing the intensity and stability of football training movements from the fused data, and then performing feature-level fusion of the posture features, temporal features, and motion dynamic features to obtain fused features for characterizing the characteristics of football training movements includes: Based on the training video data in the fused data, key body parts of the target trainee during football training are detected to obtain the spatial position information of each key body part during continuous training. Combined with the body parameter information in the target trainee's identity information, the spatial position information is normalized to obtain the posture features. Specifically, using training video data from the fused data set, key body parts of the target trainee during football training are detected. These key body parts typically include lower limb joints, key trunk nodes, and limb connection points closely related to football movements. By processing multiple consecutive frames of video images, the spatial position information of each key body part at continuous training moments can be obtained, thus reflecting the posture changes of training movements in the spatial dimension. Since different trainees differ in height, weight, limb length, etc., this step further combines body parameter information from the target trainee's identity information to normalize the aforementioned spatial position information, making the posture features comparable across different trainees. The posture features obtained in this way can accurately characterize the spatial structure and limb coordination relationships of football training movements, providing a stable spatial basis for subsequent movement analysis.

[0044] Based on the training video data and motion data in the fused data, the movement change trend of the target trainee's football training movements during continuous training is analyzed, and the movement change trend is corrected by combining the environmental state data to extract the temporal features. Specifically, by comprehensively utilizing training video data and motion data from the fused data, the patterns of movement changes over time during football training are analyzed. By tracking movement execution during continuous training sessions, the trends in the initiation, transition, and termination phases of movements can be extracted, thus reflecting the continuity and rhythmic characteristics of football training movements. Simultaneously, considering that environmental factors such as temperature and humidity reflected in the environmental data may affect the rhythm and amplitude of the trainees' movements, this step incorporates environmental data to correct the movement change trends, enabling the extracted temporal features to more accurately reflect the inherent patterns of change in the training movements themselves. This processing method helps improve the ability of temporal features to represent the continuity and stability of movements.

[0045] Based on the motion data in the fused data, the amplitude and fluctuation characteristics of the motion parameters of the target trainee during football training are analyzed, and the changes in the motion parameters are compensated by combining the environmental state data to extract the motion dynamic features. Specifically, based on motion data from the fused data, the changes in motion parameters of the target trainee during football training are analyzed. These changes include the amplitude and fluctuation characteristics of parameters such as acceleration and angular velocity during training. By analyzing the changes in these motion parameters at different training stages, the intensity level of the training movements and the stability characteristics during movement execution can be reflected. Considering that external conditions reflected in the environmental state data may affect the sensor acquisition results, such as differences in sports performance under high temperature or high humidity environments, this step incorporates environmental state data to compensate for changes in motion parameters, thereby reducing the interference of environmental factors on the results of motion dynamic feature extraction. The motion dynamic features extracted in this way can more objectively reflect the true motion state of football training movements.

[0046] The posture features, temporal features, and motion dynamic features are subjected to feature scale unification processing. After the feature scale unification processing is completed, the posture features, temporal features, and motion dynamic features are fused to generate fused features that characterize the characteristics of football training movements.

[0047] Specifically, the extracted posture features, temporal features, and motion dynamic features undergo unified processing. Since different features have different sources and physical meanings, their numerical ranges and scales often vary significantly. Therefore, it is necessary to first unify the feature scales of each type of feature to avoid any one type of feature having excessive weight in subsequent analysis. After scale unification, the posture features, temporal features, and motion dynamic features are fused at the feature level, allowing the three types of features to form a comprehensive description within the same feature space. Through this fusion process, the generated fused features can comprehensively characterize the characteristics of football training movements from multiple dimensions such as spatial posture, temporal evolution, and motion intensity, providing complete and structurally sound input features for subsequent action evaluation models.

[0048] In an optional embodiment, based on the training video data in the fused data, key body parts of the target trainee during football training are detected to obtain spatial position information of each key body part at continuous training moments. This spatial position information is then normalized by combining it with body parameter information from the target trainee's identity information to obtain the posture features, which include: The video images in each frame of the training video data are preprocessed, and the preprocessing includes at least one image denoising, image cropping or image scaling; Specifically, preprocessing is performed on each frame of the training video data. The core purpose is to unify and optimize the quality and scale of the original video images before proceeding to the subsequent recognition of key human body parts. Training videos are easily affected by factors such as lighting variations, complex backgrounds, and camera noise during actual acquisition. Directly using the original images would reduce the accuracy of subsequent recognition. Therefore, image denoising can reduce the interference of random noise on human body edges and joint areas; image cropping can focus the image on the target trainee's area, reducing the influence of irrelevant background information; and image scaling is used to unify the resolution and aspect ratio of images from different video frames or different acquisition devices, ensuring consistent input specifications for subsequent algorithm processing. After these preprocessing operations, the clarity, stability, and consistency of the video images are significantly improved, providing a reliable data foundation for the accurate recognition of key human body parts, thereby improving the stability and robustness of the overall pose analysis.

[0049] Based on the preprocessed video images of each frame, the key body parts of the target trainee are identified and located, and the two-dimensional or three-dimensional coordinate information of each key body part in the corresponding video frame is determined. Specifically, based on preprocessed video frames, key body parts of the target trainee are identified and located, obtaining their two-dimensional or three-dimensional coordinates within the corresponding video frames. Key body parts typically refer to joints or skeletal nodes closely related to soccer training movements, such as the head, shoulders, hips, knees, and ankles. These parts directly reflect spatial posture changes during training movements. In practice, computer vision-based human posture estimation algorithms can be used to analyze the human structure in each frame, automatically detecting the positions of key body parts within the image. When the training system is equipped with multiple cameras or depth cameras, the coordinates of the key body parts in three-dimensional space can be further reconstructed. By accurately acquiring the coordinate information of key body parts at the single-frame level, the system can meticulously depict the trainee's body posture at any given moment, laying the foundation for subsequent continuous motion analysis.

[0050] According to the time sequence of video frames, the coordinate information of the key body parts is arranged in time sequence to obtain the spatial position information sequence of each key body part at continuous training time. Specifically, the coordinate information of key body parts is arranged chronologically according to the video frames, forming a sequence of spatial position information for each key part during continuous training. Because football training movements are inherently continuous and dynamic, relying solely on a single frame's posture is insufficient to reflect the complete movement process. Therefore, it is necessary to organize posture information from different time points according to the video frame sequence. In implementation, the coordinates of the same key part in adjacent frames can be sequentially connected based on the timestamps or frame numbers carried by the video frames themselves, forming a spatial trajectory that changes over time. This sequence of spatial position information not only reflects the positional changes of key parts but also implicitly contains trends in the speed and direction of the movement, providing crucial information for subsequently extracting the rhythm and continuity features of the movement, thus more comprehensively describing the dynamic process of football training movements.

[0051] Obtain body parameter information related to body structure from the target trainee's identity information, wherein the body parameter information includes at least height, weight or limb length parameters; Specifically, different trainees have natural differences in height, body shape, and limb proportions. If raw spatial coordinates are used directly for analysis, individual differences can easily be misinterpreted as differences in movement. Therefore, a database of trainee identity information is pre-established within the system, and each trainee is linked to a unique identifier in the current training process. This allows the system to simultaneously acquire corresponding body parameters while analyzing posture. By introducing this information related to body structure, a basis is provided for subsequent individualized correction of posture data, helping to eliminate interference from differences in physical conditions and improving the fairness and comparability of movement evaluation.

[0052] Based on the body parameter information, the spatial position information sequence is normalized to obtain posture features that characterize the spatial posture changes of football training movements.

[0053] Specifically, normalization typically involves scaling the spatial coordinates of key body parts according to the trainee's height or limb length, or using a reference body part as the origin, thus transforming the posture representation into relative structural features independent of individual size. In this way, even when different trainees perform the same movement, their posture features exhibit high numerical consistency, facilitating unified analysis and comparison in subsequent models. Normalized posture features more realistically reflect the spatial structure and variation patterns of the movement itself, reducing the impact of individual differences and improving the generalization ability and accuracy of subsequent movement evaluation models across different trainees.

[0054] In an optional embodiment, the step of analyzing the movement change trend of the target trainee's football training movements during continuous training based on the training video data and motion data in the fused data, and correcting the movement change trend by combining the environmental state data, and extracting the temporal features includes: From the fused data, training video data and motion data corresponding to the target trainee at continuous training moments are extracted to construct a motion time sequence data sequence for characterizing the football training process. Specifically, training video data and motion data corresponding to the target trainee at continuous training moments are extracted from the fused data, and a temporal sequence of movements is constructed. The core of this approach lies in reorganizing information originally scattered across different modalities into a continuous data structure that reflects the complete training process along a unified training timeline. This temporal sequence of movements is not a simple data splicing; rather, it uses time as the main thread, mapping video posture information at each training moment to synchronous motion sensor data, ensuring that each point in time simultaneously reflects the trainee's external movement performance and internal motion state. In this way, the continuous evolutionary characteristics of football training movements over time can be fully preserved, providing a foundation for subsequent analysis of movement trends while avoiding incomplete movement understanding due to data fragmentation.

[0055] Based on the action time sequence data, action change parameters between adjacent training moments are extracted. The action change parameters include at least one of the following: joint angle change, limb displacement change, action execution rhythm change, and movement speed change. Specifically, motion variation parameters directly characterize the dynamic properties of training movements. For example, changes in joint angles reflect variations in the amplitude of limb flexion, extension, or swing; changes in limb displacement describe the movement of different body parts in space; changes in the rhythm of motion execution reflect the speed and continuity of the movement; and changes in movement speed are closely related to the explosiveness and continuity of the movement. In implementation, by performing differential calculations or trend estimations on parameters corresponding to adjacent moments, the originally continuous motion process can be transformed into a series of quantifiable variation indicators, enabling the system to perceive subtle changes in football training movements in a data-driven manner.

[0056] Based on the motion change parameters, analyze the change trend of the target trainee's football training motion under the continuous training time to obtain the initial motion temporal characteristics; Specifically, the trend of movement change emphasizes whether, over a period of training, the movement tends to stabilize and gradually accelerate, or exhibits significant fluctuations or discontinuities. In practical analysis, time series analysis, sliding statistics, or trend fitting of movement change parameters can be used to extract feature information reflecting the overall pattern of change, rather than being limited to a single instantaneous state. The resulting initial movement temporal features can comprehensively reflect the continuity, rhythmic consistency, and evolutionary patterns of football training movements over time, elevating the system's understanding of training movements from a static posture level to a dynamic process level.

[0057] Based on the environmental state data, environmental impact correction processing is performed on the initial action timing characteristics to determine the timing characteristics; Specifically, football training often takes place under varying environmental conditions. For example, changes in temperature and humidity can affect a trainee's physical condition and movement rhythm. Without differentiation, changes caused by environmental factors can easily be misjudged as problems with movement quality. Therefore, by introducing environmental state data corresponding to the training period and correcting the initial movement timing characteristics, the objective influence of the external environment can be comprehensively considered when analyzing movement trends. In practice, the amplitude or rhythm characteristics of movement changes can be weighted and adjusted or compensated based on different environmental parameters, making the resulting timing characteristics closer to the trainee's performance under real conditions. This not only improves the objectivity of the timing characteristics but also enhances the consistency and reliability of subsequent movement evaluation results across different training environments.

[0058] The process involves analyzing the amplitude and fluctuation characteristics of movement parameters of the target trainee during football training based on the motion data in the fused data, and compensating for the changes in the movement parameters by combining the environmental state data, thereby extracting the dynamic features of the movement, including: Based on the motion data, the variation amplitude and fluctuation characteristics of each motion parameter of the target trainee during continuous training are calculated to obtain the initial motion dynamic parameters; Specifically, the motion data here mainly comes from parameters such as acceleration and angular velocity collected by wearable inertial sensors. These parameters directly reflect the speed, force, and changes in the trainee's limb movements. During implementation, the system compares and analyzes motion parameters at consecutive points along a unified training timeline. For example, it calculates the numerical difference and rate of change between adjacent sampling points, as well as the mean, variance, or extreme value range within a certain time window, thereby characterizing the overall amplitude and fluctuation of motion parameters during training. The initial dynamic motion parameters obtained in this way can objectively reflect the dynamic characteristics of football training movements during execution, laying a data foundation for subsequent judgments of movement intensity and stability.

[0059] Based on the initial motion dynamic parameters, analyze the changes in motion intensity and motion stability characteristics of the target trainee's football training movements during the training process; Specifically, changes in exercise intensity are mainly reflected in the magnitude of the amplitude of motion parameters and their trends over time. For example, the peak value of acceleration can reflect the explosiveness of the movement, while the range of angular velocity variation is closely related to the amplitude and speed of the movement. Movement stability is more about whether the fluctuations in motion parameters are smooth, and whether there are frequent large fluctuations or abnormal sudden changes. In specific analysis, the distribution characteristics of the initial dynamic parameters during training can be statistically analyzed, and phased comparisons or trend analyses can be performed to distinguish whether the movements are continuous and whether there are obvious imbalances or instabilities. Through this analysis process, the system can extract dynamic characteristic information highly correlated with the quality of football training movements from a purely data-driven perspective.

[0060] Based on the environmental state data, environmental compensation processing is performed on the initial motion dynamic parameters to determine the motion dynamic features.

[0061] Specifically, since training environmental conditions such as temperature and humidity can affect a trainee's energy expenditure, range of motion, and stability, directly using initial motion dynamic parameters could easily misinterpret changes caused by environmental factors as movement problems. Therefore, in implementation, the system incorporates environmental state data from the corresponding training period to correct the initial motion dynamic parameters. For example, in high-temperature or high-humidity environments, the system appropriately compensates for or adjusts the fluctuation range of motion parameters, making the analysis results more consistent with the trainee's actual performance under those conditions. The motion dynamic characteristics after environmental compensation can more accurately reflect the intensity and stability of the training movement itself, thereby improving the objectivity and reliability of subsequent movement evaluation and optimization suggestions.

[0062] In an optional embodiment, the step of performing feature scale unification processing on the posture features, the temporal features, and the motion dynamic features, and then fusing the posture features, the temporal features, and the motion dynamic features after completing the feature scale unification processing to generate fused features for characterizing the characteristics of football training movements includes: The posture features, temporal features, and motion dynamic features are subjected to feature scale unification processing to obtain scale-unified posture features, temporal features, and motion dynamic features; Specifically, standardizing the feature scale of posture features, temporal features, and motion dynamic features aims to address the differences in numerical range, dimensions, and statistical distribution among different types of features. Posture features typically originate from the spatial location or angular information of key body points, and their numerical range may be related to human body dimensions. Temporal features often manifest as rates of change, trend parameters, or time-related indicators. Motion dynamic features are mostly physical quantities such as acceleration and angular velocity, and their numerical fluctuation range is closely related to sensor sampling accuracy and motion intensity. In the implementation process, the system can normalize or standardize each type of feature separately, for example, by linear scaling based on minimum and maximum values, or by standardization transformation based on mean and standard deviation, so that different features are expressed within a unified numerical range. This process prevents one type of feature from dominating the overall result due to excessively large numerical amplitudes during subsequent fusion, thus ensuring the comparability of multimodal features within the same evaluation framework.

[0063] Based on the target trainee's identity information and the environmental status data, the feature weight parameters of various features in the evaluation of football training actions are determined. Specifically, factors such as body parameters, training level, and age included in the target trainee's identity information can influence the importance of different features in movement evaluation. For example, for trainees with significant differences in height and limb length, the weights of posture features need to be adjusted appropriately. Environmental data such as temperature and humidity may affect movement stability and exercise intensity, thus influencing the role of dynamic movement features in the overall evaluation. In implementation, the system can establish a mapping relationship between identity information, environmental conditions, and feature weights based on pre-defined rule models or through statistical analysis of historical training data. This allows for the dynamic generation of posture feature weights, temporal feature weights, and dynamic movement feature weights for the corresponding training scenario, making the weight allocation more closely aligned with actual training conditions.

[0064] Based on the feature weight parameters, the scale-unified pose features, temporal features, and motion dynamic features are weighted and fused to generate initial fused features; Specifically, each type of feature is multiplied by its corresponding weight coefficient, and then combined according to a unified time index or feature dimension to form a comprehensive feature representation containing information from multiple types of features. This weighted fusion process not only preserves the discriminative information of each type of feature, but also reflects the importance of different features in the current trainee and training environment through weight adjustment, enabling the initial fused features to more comprehensively and reasonably represent the overall characteristics of football training movements.

[0065] The initial fused features are processed by feature concatenation, dimensionality reduction, or feature selection to obtain the fused features.

[0066] Specifically, since the initial fused features may have high dimensionality, and there may be redundancy or correlation between different features, the implementation process can employ feature concatenation to form a unified feature vector, combined with dimensionality reduction methods to compress high-dimensional features, or feature selection algorithms to retain the subset of features most discriminative for action evaluation. This process reduces the computational complexity of subsequent deep learning models and highlights key information closely related to the quality of football training movements, thereby improving the stability and accuracy of action evaluation results.

[0067] In an optional embodiment, the fused features are input into a pre-trained deep learning-based football training action evaluation model for inference analysis, and the model outputs action evaluation results corresponding to the football training actions. The action evaluation results include action standardization scores and action completion quality scores, including: The fused features are processed to conform to the input dimension and data structure requirements of the football training action evaluation model. Specifically, the core purpose of formatting the fused features is to ensure that the multi-source fused feature data can be stably and accurately received and processed by the subsequent football training action evaluation model. Fusion features are typically composed of multiple types of features, such as posture, temporal sequence, and motion dynamics. During generation, inconsistencies in feature dimensions, time lengths, or data arrangement order may exist. Therefore, in implementation, the fused features need to be structured according to the input specifications of the pre-trained model. For example, features may be segmented or padded according to a fixed time step, features of different dimensions may be uniformly arranged into vector or tensor forms, and missing or outlier values ​​may be padded or smoothed. This process ensures that the fused features meet the model design requirements in terms of the number of dimensions, data type, and storage structure, thereby avoiding inference biases caused by non-standard inputs and improving the reliability of subsequent action evaluation analysis.

[0068] The fused features are then input into the pre-trained football training action evaluation model based on a deep learning architecture, and forward inference calculations are performed to obtain intermediate action evaluation feature representations corresponding to the football training actions. Specifically, the core step in the entire action evaluation process is to input the processed fused features into a pre-trained deep learning-based football training action evaluation model for forward inference computation. This model is typically built on a deep neural network, capable of multi-level feature abstraction and representation of complex spatiotemporal features. In its implementation, the fused features serve as model input, undergoing sequential processing through multiple network layers, including feature mapping, nonlinear transformations, and cross-time or cross-feature dimension correlation modeling, thereby gradually extracting high-level semantic information closely related to the quality of football training actions. Through forward inference computation, the model can transform the original fused features into a more abstract and discriminative intermediate action evaluation feature representation, providing a unified data foundation for subsequent analysis across different evaluation dimensions.

[0069] Based on the intermediate action evaluation feature representation, a normative evaluation branch for characterizing the standardization of football training actions and a quality evaluation branch for characterizing the completion quality of football training actions are constructed respectively. Specifically, constructing normative and quality evaluation branches based on intermediate motion evaluation feature representations is to achieve targeted analysis of different evaluation dimensions of football training movements. The intermediate motion evaluation feature representations already comprehensively reflect the overall characteristics of training movements at the temporal, spatial, and dynamic levels. Building upon this, by setting different analysis branches within the model, the same feature representation can be mapped to different evaluation objectives. The normative evaluation branch focuses on analyzing whether the movement conforms to standard technical requirements and training specifications, while the quality evaluation branch focuses more on the coherence, stability, and overall performance level during the movement's completion. In implementation, these two branches can employ different network parameters or structural forms to model intermediate features differentially. This allows the model to capture key features of the degree of motion standardization and the quality of motion completion while maintaining overall consistency, thus laying the foundation for subsequent multi-dimensional evaluation results.

[0070] The intermediate action evaluation feature representation is analyzed through the normative evaluation branch, and the action normative score corresponding to the football training action is output. Specifically, the standardization evaluation branch analyzes the intermediate movement evaluation feature representations and outputs a movement standardization score, focusing on quantitatively judging whether football training movements conform to established technical standards. The standardization evaluation branch typically analyzes the standard trajectory, key posture positions, and movement sequence of the movement. In practice, it matches or compares the intermediate movement evaluation features with the standard movement feature patterns implicitly learned in the model, calculating the degree of deviation between the current training movement and the standard movement. This scoring process not only reflects the standardization of individual movements in local aspects but also comprehensively assesses the conformity of the entire set of movements in terms of temporal continuity and technical integrity, thus forming a score result that intuitively reflects the level of movement standardization, helping trainees quickly identify non-standard elements in the movements.

[0071] The intermediate action evaluation feature representation is analyzed through the quality evaluation branch, and the action completion quality score corresponding to the football training action is output. Specifically, the quality evaluation branch analyzes the intermediate movement evaluation features and outputs a movement completion quality score, focusing on the overall performance of football training movements during actual execution. Unlike standardization evaluation, quality evaluation pays more attention to the fluidity, stability, and rationality of force and rhythm of movements. In implementation, the quality evaluation branch comprehensively analyzes the movement execution process based on the information on movement intensity changes, temporal consistency, and dynamic stability contained in the intermediate movement evaluation features, thereby judging whether the movement is continuous and whether there are obvious fluctuations or interruptions. The movement completion quality score obtained in this way can reflect the completion level of training movements from an overall performance perspective, providing trainees with more practically instructive evaluation results.

[0072] The action standardization score and the action completion quality score are combined to form the action evaluation result of the football training action.

[0073] Specifically, combining the performance evaluation scores for both movement standardization and execution quality to form the final evaluation result for football training movements represents a unified integration of multi-dimensional evaluation information. In practice, the two types of scores can be weighted or comprehensively mapped according to preset evaluation rules or weighting relationships to obtain an evaluation result that comprehensively reflects the level of training movements. This combination process retains the individual evaluation value of both movement standardization and execution quality while avoiding the bias caused by a single indicator, resulting in a more comprehensive and objective final evaluation result. In this way, trainees and coaches can make an overall judgment on training movements based on a unified evaluation result, providing a clear and reliable basis for generating subsequent movement optimization suggestions. Example 2

[0074] In addition, combined Figure 1 The football training motion evaluation and optimization method that integrates multimodal data described in Embodiment 1 of the present invention can be implemented by a football training motion evaluation and optimization system that integrates multimodal data. Figure 3 The diagram shows a hardware structure of the football training action evaluation and optimization system that integrates multimodal data, as provided in Embodiment 2 of the present invention.

[0075] A football training motion evaluation and optimization system that integrates multimodal data may include a processor and a memory storing computer program instructions.

[0076] Specifically, the processor may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement embodiments of the present invention.

[0077] The memory may include a large-capacity storage device for data or instructions. For example, and not limitingly, the memory may include a hard disk drive (HDD), a floppy disk drive, flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory may include removable or non-removable (or fixed) media. Where appropriate, the memory may be internal or external to a data processing device. In a particular embodiment, the memory is a non-volatile solid-state memory. In a particular embodiment, the memory includes a read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0078] The processor reads and executes computer program instructions stored in memory to implement any of the football training action evaluation and optimization methods that integrate multimodal data in the above embodiments.

[0079] In one example, a football training motion evaluation and optimization system that integrates multimodal data may also include a communication interface and a bus. For example, Figure 3 As shown, the processor, memory, and communication interface are connected via a bus and communicate with each other.

[0080] The communication interface is mainly used to enable communication between various modules, devices, units and / or equipment in the embodiments of the present invention.

[0081] A bus, including hardware, software, or both, couples components of the device together. For example, and not limitingly, a bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, a bus may include one or more buses. While specific buses are described and illustrated in embodiments of the invention, the invention contemplates any suitable bus or interconnect.

[0082] In summary, the embodiments of the present invention provide a method and system for evaluating and optimizing football training movements by integrating multimodal data.

[0083] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0084] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the required tasks. The programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0085] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant locality, and corresponding operation entry points shall be provided for the user to choose to authorize or refuse.

[0086] It should also be noted that the exemplary embodiments mentioned in this invention describe methods or systems based on a series of steps or apparatus. However, this invention is not limited to the order of the steps described above; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0087] The above description is merely a specific embodiment of the present invention. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention.

Claims

1. A method for evaluating and optimizing football training movements by integrating multimodal data, characterized in that, The method includes: Acquire training video data, motion data, trainee identity data, and environmental data of the target trainee during football training, wherein the motion data includes motion parameter data collected by wearable sensors; Based on the timestamp information of the training video data and the motion data, the training video data, the motion data, the trainee identity data and the environmental data are time-synchronized and aligned, and then fused after time alignment to obtain fused data corresponding to the football training process. The posture features used to characterize the spatial posture changes of football training movements, the temporal features used to characterize the continuous change law of football training movements, and the motion dynamic features used to characterize the intensity and stability of football training movements are extracted from the fused data respectively. The posture features, the temporal features, and the motion dynamic features are then fused at the feature level to obtain the fused features used to characterize the characteristics of football training movements. The fused features are input into a pre-trained football training action evaluation model based on a deep learning architecture for inference analysis, and the action evaluation results corresponding to the football training actions are output. The action evaluation results include action standardization score and action completion quality score. Based on the action evaluation results, action optimization suggestions are generated corresponding to the football training actions. These suggestions are used to guide the target trainee to adjust their football training actions.

2. The method for evaluating and optimizing football training movements by fusing multimodal data according to claim 1, characterized in that, The acquisition of training video data, movement data, trainee identity data, and environmental data of the target trainee during football training includes: The training movements of the target trainee during football training are filmed to obtain a series of multiple video images that record the football training process of the target trainee. The video images are then processed by frame serialization to obtain the training video data corresponding to the training. Wearable inertial sensors are fixedly worn on the lower limbs of the target trainee. During football training, the three-axis acceleration data and three-axis angular velocity data of the target trainee are collected at a preset sampling frequency. The collected data are formatted to obtain the corresponding motion parameter data as the motion data. Based on the unique identification information corresponding to the target trainee, the trainee identity data corresponding to the target trainee is read from the pre-established trainee information database; By installing an environmental monitoring module at the football training field, environmental parameters during football training are collected to obtain the environmental data. The environmental parameters include at least ambient temperature and ambient humidity.

3. The method for evaluating and optimizing football training movements by fusing multimodal data according to claim 1, characterized in that, The process of synchronizing and aligning the training video data, the exercise data, the trainee identity data, and the environmental data based on the timestamp information of the training video data and the exercise data, and then fusing them after time alignment, to obtain fused data corresponding to the football training process includes: Extract the first timestamp information corresponding to each frame of video image in the training video data, and extract the second timestamp information corresponding to each sampling point in the motion data; Based on the first timestamp information and the second timestamp information, a unified training timeline for characterizing the football training process is constructed. According to the unified training timeline, the training video data and the motion data are subjected to time mapping processing so that each frame of video image and the corresponding motion data at the time dimension are established. When the sampling frequencies of the training video data and the motion data are inconsistent on the unified training time axis, interpolation or resampling is performed on the data with the lower sampling frequency to achieve synchronous alignment of the training video data and the motion data in the time dimension. The training video data and motion data that have been synchronized and aligned in time are associated and fused with the trainee identity data and environmental data within the corresponding training time period to obtain fused data corresponding to the football training process.

4. The method for evaluating and optimizing football training movements by fusing multimodal data according to claim 3, characterized in that, The process of associating and fusing the training video data and motion data that have been synchronized and aligned in time with the trainee identity data and environmental data within the corresponding training time period to obtain fused data corresponding to the football training process includes: Based on the unified training timeline, the target training time period corresponding to the training video data and the motion data that have completed time synchronization alignment is determined; The target training time period is matched with the trainee identity data to determine the target trainee identity information corresponding to the target training time period; Based on the target training time period, filter environmental state data corresponding to the target training time period from the environmental data; Using the unified training timeline as a time index, the training video data, the motion data, the target trainee identity information, and the environmental state data are organized to construct a multimodal data structure corresponding to the football training process; The multimodal data structure is fused to generate fused data corresponding to the football training process.

5. The method for evaluating and optimizing football training movements by fusing multimodal data according to claim 4, characterized in that, The process involves extracting posture features representing spatial changes in football training movements, temporal features representing continuous changes in football training movements, and dynamic features representing the intensity and stability of football training movements from the fused data. Then, the posture features, temporal features, and dynamic features are fused at the feature level to obtain fused features representing the characteristics of football training movements, including: Based on the training video data in the fused data, key body parts of the target trainee during football training are detected to obtain the spatial position information of each key body part during continuous training. Combined with the body parameter information in the target trainee's identity information, the spatial position information is normalized to obtain the posture features. Based on the training video data and motion data in the fused data, the movement change trend of the target trainee's football training movements during continuous training is analyzed, and the movement change trend is corrected by combining the environmental state data to extract the temporal features. Based on the motion data in the fused data, the amplitude and fluctuation characteristics of the motion parameters of the target trainee during football training are analyzed, and the changes in the motion parameters are compensated by combining the environmental state data to extract the motion dynamic features. The posture features, temporal features, and motion dynamic features are subjected to feature scale unification processing. After the feature scale unification processing is completed, the posture features, temporal features, and motion dynamic features are fused to generate fused features that characterize the characteristics of football training movements.

6. The method for evaluating and optimizing football training movements by fusing multimodal data according to claim 5, characterized in that, The training video data from the fused data is used to detect key body parts of the target trainee during football training, obtaining spatial position information of each key body part during continuous training. This spatial position information is then normalized by combining it with body parameter information from the target trainee's identity information to obtain the posture features, including: The video images in each frame of the training video data are preprocessed, and the preprocessing includes at least one image denoising, image cropping or image scaling; Based on the preprocessed video images of each frame, the key body parts of the target trainee are identified and located, and the two-dimensional or three-dimensional coordinate information of each key body part in the corresponding video frame is determined. According to the time sequence of video frames, the coordinate information of the key body parts is arranged in time sequence to obtain the spatial position information sequence of each key body part at continuous training time. Obtain body parameter information related to body structure from the target trainee's identity information, wherein the body parameter information includes at least height, weight or limb length parameters; Based on the body parameter information, the spatial position information sequence is normalized to obtain posture features that characterize the spatial posture changes of football training movements.

7. The method for evaluating and optimizing football training movements by fusing multimodal data according to claim 5, characterized in that, The method involves analyzing the movement trends of the target trainee's soccer training actions over continuous training sessions based on the training video data and motion data from the fused data, and then correcting these movement trends using the environmental state data. The extraction of the temporal features includes: From the fused data, training video data and motion data corresponding to the target trainee at continuous training moments are extracted to construct a motion time sequence data sequence for characterizing the football training process. Based on the action time sequence data, action change parameters between adjacent training moments are extracted. The action change parameters include at least one of the following: joint angle change, limb displacement change, action execution rhythm change, and movement speed change. Based on the motion change parameters, analyze the change trend of the target trainee's football training motion under the continuous training time to obtain the initial motion temporal characteristics; Based on the environmental state data, environmental impact correction processing is performed on the initial action timing characteristics to determine the timing characteristics; The process involves analyzing the amplitude and fluctuation characteristics of movement parameters of the target trainee during football training based on the motion data in the fused data, and compensating for the changes in the movement parameters by combining the environmental state data, thereby extracting the dynamic features of the movement, including: Based on the motion data, the variation amplitude and fluctuation characteristics of each motion parameter of the target trainee during continuous training are calculated to obtain the initial motion dynamic parameters; Based on the initial motion dynamic parameters, analyze the changes in motion intensity and motion stability characteristics of the target trainee's football training movements during the training process; Based on the environmental state data, environmental compensation processing is performed on the initial motion dynamic parameters to determine the motion dynamic features.

8. The method for evaluating and optimizing football training movements by fusing multimodal data according to claim 5, characterized in that, The process of unifying the feature scales of the posture features, temporal features, and motion dynamic features, followed by feature fusion of these features to generate fused features characterizing the characteristics of football training movements, includes: The posture features, temporal features, and motion dynamic features are subjected to feature scale unification processing to obtain scale-unified posture features, temporal features, and motion dynamic features; Based on the target trainee's identity information and the environmental status data, the feature weight parameters of various features in the evaluation of football training actions are determined. Based on the feature weight parameters, the scale-unified pose features, temporal features, and motion dynamic features are weighted and fused to generate initial fused features; The initial fused features are processed by feature concatenation, dimensionality reduction, or feature selection to obtain the fused features.

9. The method for evaluating and optimizing football training movements by fusing multimodal data according to any one of claims 1-8, characterized in that, The fused features are input into a pre-trained football training action evaluation model based on a deep learning architecture for inference analysis, and the model outputs action evaluation results corresponding to the football training actions. The action evaluation results include action standardization scores and action completion quality scores, including: The fused features are processed to conform to the input dimension and data structure requirements of the football training action evaluation model. The fused features are then input into the pre-trained football training action evaluation model based on a deep learning architecture, and forward inference calculation is performed to obtain the intermediate action evaluation feature representation corresponding to the football training action. Based on the intermediate action evaluation feature representation, a normative evaluation branch for characterizing the standardization of football training actions and a quality evaluation branch for characterizing the completion quality of football training actions are constructed respectively. The intermediate action evaluation feature representation is analyzed through the normative evaluation branch, and the action normative score corresponding to the football training action is output. The intermediate action evaluation feature representation is analyzed through the quality evaluation branch, and the action completion quality score corresponding to the football training action is output. The action standardization score and the action completion quality score are combined to form the action evaluation result of the football training action.

10. A football training motion evaluation and optimization system integrating multimodal data, characterized in that, include: At least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the method as described in any one of claims 1-9.

Images