Basketball skill motion intelligent evaluation method and system based on acoustic characteristics

Abstract

The invention relates to the technical field of sports skill analysis, and provides a basketball skill action intelligent evaluation method and system based on acoustic characteristics. The method comprises the following steps: acquiring and decomposing acoustic signals to extract a feature set; positioning a technical action event time interval based on the energy abrupt change point and the spectrum mode; matching the acoustic feature subset to determine a target technical action type; extracting quality evaluation features and comparing the quality evaluation features with a standard acoustic feature reference to determine a quality score; and associating the action type with the score to generate an evaluation result. According to the invention, objective assessment of basketball skilled actions is realized, and the accuracy and efficiency of assessment are improved.

Description

Technical Field

[0001] This invention relates to the field of sports technology analysis, and in particular to a method and system for intelligent evaluation of basketball technical movements based on acoustic features. Background Technology

[0002] With the continuous development of sports technology, basketball technique analysis and evaluation has become a crucial aspect of improving athletes' training effectiveness and competitive performance. Traditional basketball technique evaluation primarily relies on coaches' subjective judgment and video analysis, methods often limited by the accuracy and consistency of human observation. In recent years, various sensing technologies have been introduced into the field of basketball technique analysis, including visual recognition systems, inertial measurement units, and wearable devices. These technologies can capture and quantitatively analyze athletes' movement characteristics. However, acoustic signals, as a non-contact information acquisition method, contain rich information about movement, and their application potential in basketball technique evaluation has not yet been fully explored. Various sounds generated during basketball, such as dribbling sounds, shooting sounds, and footsteps, contain important information about the quality of technique execution and can serve as a unique dimension for evaluating technique level.

[0003] The main deficiencies and shortcomings in the current field of basketball technical movement assessment include: Existing methods for evaluating basketball skills rely excessively on visual information and inertial sensor data, neglecting the rich technical movement characteristics contained in acoustic signals. This results in a single evaluation dimension, making it difficult to fully capture the subtle changes and quality differences in movements.

[0004] Traditional evaluation systems struggle to pinpoint the precise time boundaries of basketball technical movements, especially in the context of continuous and rapidly changing basketball scenarios. They are unable to accurately identify and segment different technical movement events, affecting the accuracy and relevance of subsequent evaluations.

[0005] Current technology lacks objective and quantitative standards and methods for evaluating the quality of basketball technical movements. Evaluation results are often influenced by subjective factors, and there are significant differences among different evaluators, making it difficult to provide athletes with consistent technical improvement guidance and training feedback. Summary of the Invention

[0006] The present invention provides a method and system for intelligent evaluation of basketball technical movements based on acoustic features, which can at least solve some of the problems existing in the prior art.

[0007] A first aspect of this invention provides an intelligent evaluation method for basketball technical movements based on acoustic features, comprising: Acoustic signals generated during basketball play are acquired, and the acoustic signals are decomposed in the time-frequency domain to extract a set of acoustic features. Based on the energy mutation points and spectral patterns in the acoustic feature set, the time boundaries of multiple technical action events in basketball are located to obtain the time interval of each technical action event; For each technical action event within a given time interval, an acoustic feature subset is extracted. Based on this subset, a matching operation is performed between the acoustic feature subset and the acoustic feature representations of each basketball technical action type to determine the target technical action type corresponding to the technical action event. For the target technical action type, quality evaluation features associated with the standard execution mode of the target technical action type are extracted from the acoustic feature subset, and a quality score for the technical action event is determined based on the deviation between the quality evaluation features and a preset standard acoustic feature benchmark for the target technical action type. The target technical movement type and the quality score are correlated to generate an evaluation result for basketball technical movements.

[0008] Based on the energy abrupt change points and spectral patterns in the acoustic feature set, the temporal boundaries of multiple technical action events in basketball are located to obtain the time intervals of each technical action event, including: Time series analysis is performed on the energy distribution characteristics in the acoustic feature set. By calculating the energy change rate between adjacent time windows, time points where the energy change rate exceeds a dynamic threshold are identified as energy mutation points. The dynamic threshold is adaptively adjusted based on the background energy level and energy fluctuation amplitude of the acoustic signal. For each energy mutation point, extract the spectral envelope features within a preset time range before and after the energy mutation point; The spectral envelope features are compared with the pre-stored spectral feature benchmarks for the start and end phases of various basketball techniques. The energy mutation point that matches the spectral feature benchmark for the start phase of the technique is taken as the start boundary point of the technique, and the energy mutation point that matches the spectral feature benchmark for the end phase of the technique is taken as the end boundary point of the technique. Based on the time continuity constraint of basketball technical movements, the starting boundary point and the ending boundary point of the movement that satisfy the continuity constraint in time sequence are paired to obtain the time interval of each technical movement event.

[0009] For each technical action event within a given time interval, an acoustic feature subset is extracted. Based on this subset, a matching operation is performed between the acoustic feature subset and the acoustic feature representations of various basketball technical action types to determine the target technical action type corresponding to the technical action event, including: For each technical action event within a given time interval, the energy distribution within that time interval is extracted from the acoustic feature set. For each basketball technique, acoustic characteristic description parameters are extracted based on the differences in acoustic signals generated at different stages. These parameters are then combined to obtain the acoustic feature representation of the basketball technique. The acoustic feature subset is decomposed into segmented acoustic features corresponding to different stages, and the similarity between each segmented acoustic feature and the acoustic characteristic description parameters of the corresponding stage in the acoustic feature representation of each basketball movement technique is calculated. Weighted fusion of similarities at different stages yields the comprehensive matching degree between the acoustic feature subset and various basketball technique types. The basketball technique type with the highest overall matching degree is selected as the target technique type corresponding to the technique event.

[0010] For each basketball technique, based on the differences in acoustic signals generated at different stages, acoustic characteristic description parameters for each stage are extracted. These parameters are then combined to obtain the acoustic feature representation of the basketball technique, including: For each type of basketball technique, based on the physical execution process of that technique, it is divided into multiple consecutive execution stages. For each execution stage of each basketball technique, multiple sets of standard execution acoustic samples are collected for that execution stage. These standard execution acoustic samples are derived from the acoustic signal records generated at each execution stage of that basketball technique under standard execution conditions. For each execution stage, a time-frequency domain analysis is performed on the standard execution acoustic samples to extract acoustic characteristic descriptive parameters that characterize the distribution of acoustic signals in the energy domain, frequency domain, and time domain of that execution stage. The acoustic characteristic description parameters extracted from multiple sets of standard execution acoustic samples in the same execution phase are statistically aggregated to obtain representative acoustic characteristic description parameters for that execution phase. The representative acoustic characteristic parameters of each execution stage of the basketball technique are combined according to the time sequence of the execution stages to obtain the acoustic feature representation of the basketball technique.

[0011] For the target technical action type, quality evaluation features associated with the standard execution mode of the target technical action type are extracted from the acoustic feature subset. Based on the deviation between the quality evaluation features and a preset standard acoustic feature benchmark for the target technical action type, a quality score for the technical action event is determined, including: Identify multiple quality evaluation dimensions for assessing the execution quality of the target technical action type; For each quality evaluation dimension, extract the quality evaluation features associated with that quality evaluation dimension from the subset of acoustic features; For each quality evaluation dimension, a preset standard acoustic feature benchmark for the target technical action type on that quality evaluation dimension is obtained, and the deviation between the quality evaluation feature corresponding to that quality evaluation dimension and the standard acoustic feature benchmark is calculated. Based on the deviation amount of each quality evaluation dimension, a deviation vector is determined. The deviation vector is then matched with the deviation vector templates corresponding to multiple quality levels of the target technical action type in a pre-stored manner to identify the quality level with the highest similarity to the deviation vector. The quality score of the technical action event is determined based on the similarity between the deviation vector and the deviation vector template corresponding to the quality level, and the difference in similarity between the deviation vector and the deviation vector templates corresponding to adjacent quality levels.

[0012] For each quality evaluation dimension, a preset standard acoustic feature benchmark for the target technical action type on that quality evaluation dimension is obtained, and the deviation between the quality evaluation feature corresponding to that quality evaluation dimension and the standard acoustic feature benchmark is calculated, including: For the target technical action type, acoustic samples of the target technical action type at multiple different execution quality levels are collected in advance, and each acoustic sample is labeled with its corresponding execution quality level; The acoustic samples are grouped into a standard sample set according to the performance quality level. For each quality evaluation dimension, acoustic samples with a preset excellent performance quality level are selected from the standard sample set as benchmark samples. Acoustic feature analysis is performed on the benchmark sample, and the acoustic feature statistics corresponding to the quality evaluation dimension are extracted as the standard acoustic feature benchmark for the quality evaluation dimension. The initial deviation between the quality evaluation feature corresponding to the quality evaluation dimension and the standard acoustic feature benchmark is calculated. Based on the positional relationship between the quality evaluation features corresponding to the quality evaluation dimension and the acoustic feature distribution of acoustic samples with different performance quality levels in the standard sample set on the quality evaluation dimension, the deviation correction coefficient of the quality evaluation dimension is determined. The initial deviation of the quality evaluation dimension is calculated by combining it with the deviation correction coefficient to obtain the deviation of the quality evaluation dimension.

[0013] The target technical movement type and the quality score are correlated to generate an evaluation result for basketball technical movements, including: Based on the score range in which the quality score falls, the quality score is mapped to a corresponding quality level identifier, wherein the quality level identifier represents the execution quality level of the technical action event under the target technical action type; Based on the target technical action type and the deviation of each quality evaluation dimension, the quality evaluation dimensions with deviations exceeding the preset deviation threshold are identified as dimensions to be improved. For each dimension to be improved, the correlation between the dimension to be improved and the execution technical elements of the target technical action type is obtained, and improvement suggestion information for the corresponding execution technical elements is generated based on the correlation. The target technical movement type, the quality score, the quality level identifier, and the improvement suggestion information are linked and organized to generate an evaluation result for basketball technical movements.

[0014] A second aspect of the present invention provides an intelligent evaluation system for basketball technical movements based on acoustic features, comprising: The first unit is used to acquire acoustic signals generated during basketball movements, perform time-frequency domain decomposition on the acoustic signals, and extract a set of acoustic features. The second unit is used to locate the time boundaries of multiple technical action events in basketball based on the energy mutation points and spectral patterns in the acoustic feature set, and to obtain the time interval of each technical action event. The third unit is used to extract a subset of acoustic features within the time interval of each technical action event, and perform matching operations between the subset of acoustic features and the acoustic feature representations of each basketball technical action type to determine the target technical action type corresponding to the technical action event. The fourth unit is used to extract quality evaluation features associated with the standard execution mode of the target technical action type from the acoustic feature subset for the target technical action type, and to determine the quality score of the technical action event based on the deviation between the quality evaluation features and the preset standard acoustic feature benchmark of the target technical action type. The fifth unit is used to associate the target technical action type with the quality score to generate an evaluation result for basketball technical actions.

[0015] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0017] This invention achieves intelligent evaluation of basketball technical movements through acoustic signal analysis, avoiding the limitations of traditional visual analysis methods that are affected by lighting, angle, and occlusion, and providing a more reliable way to recognize technical movements.

[0018] This invention, based on time-frequency domain decomposition of acoustic features and energy mutation point analysis, can accurately locate the time boundaries of various technical action events and accurately identify the action type through feature matching operations, thus greatly improving the accuracy and stability of basketball technical action recognition.

[0019] This invention achieves an objective score for the quality of basketball technical movements by extracting quality evaluation features associated with standard execution modes and calculating the deviation from standard acoustic feature benchmarks. This provides athletes with precise basis for improving technical movements and helps to enhance training effectiveness and competitive level. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the intelligent evaluation method for basketball technical movements based on acoustic features, according to an embodiment of the present invention.

[0021] Figure 2 A schematic diagram of the acoustic feature representation process for determining the type of basketball technique movement according to an embodiment of the present invention. Detailed Implementation

[0022] 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. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0024] Figure 1 This is a flowchart illustrating the intelligent evaluation method for basketball technical movements based on acoustic features, according to an embodiment of the present invention. Figure 1 As shown, the method includes: Acoustic signals generated during basketball play are acquired, and the acoustic signals are decomposed in the time-frequency domain to extract a set of acoustic features. Based on the energy mutation points and spectral patterns in the acoustic feature set, the time boundaries of multiple technical action events in basketball are located to obtain the time interval of each technical action event; For each technical action event within a given time interval, an acoustic feature subset is extracted. Based on this subset, a matching operation is performed between the acoustic feature subset and the acoustic feature representations of each basketball technical action type to determine the target technical action type corresponding to the technical action event. For the target technical action type, quality evaluation features associated with the standard execution mode of the target technical action type are extracted from the acoustic feature subset, and a quality score for the technical action event is determined based on the deviation between the quality evaluation features and a preset standard acoustic feature benchmark for the target technical action type. The target technical movement type and the quality score are correlated to generate an evaluation result for basketball technical movements.

[0025] This invention can identify and assess the quality of technical movements by collecting and analyzing the sound signals generated during basketball games.

[0026] In the intelligent evaluation of basketball technical movements, the first step is to acquire the acoustic signals generated during the basketball game. This can be done by installing multiple high-precision microphones around the basketball court. These microphones are set to a sampling frequency of 44.1kHz and a quantization precision of 16 bits to ensure that the acquired sound signals have sufficient detail. The acquisition devices transmit the sound data to a processing server via a wireless network, forming the raw acoustic signal data stream. For example, in a basketball practice session, five microphones can be installed at the four corners of the court and near the basketball hoop to form a sound acquisition array, ensuring comprehensive coverage of sound events within the court.

[0027] Time-frequency domain decomposition of the acquired acoustic signal is a crucial step in extracting effective features. A short-time Fourier transform (SFT) method is used to process the acoustic signal, with a window size of 1024 sampling points and an overlap rate of 50%, achieving a good balance between time and frequency resolution. Through time-frequency decomposition, the sound signal is converted into a time-spectrum representation. Based on the time-spectrum, an acoustic feature set is extracted, including Mel-frequency cepstral coefficients, spectral centroid, spectral flux, zero-crossing rate, and energy envelope. For basketball dribbling, the characteristic is a regular energy change within the 20-200Hz range, while a brief high-energy peak appears in the 500-2000Hz range during a successful shot. From each time frame, 20-dimensional Mel-frequency cepstral coefficients, 5-dimensional spectral statistical features, and time-domain energy features are extracted to form a comprehensive acoustic feature set.

[0028] After acquiring the acoustic feature set, it is necessary to determine the temporal boundaries of each technical action event. This can be achieved by analyzing energy abrupt changes and spectral pattern changes in the acoustic feature set. This invention designs an adaptive threshold detection algorithm. This algorithm calculates the first-order difference of short-time energy. When the difference value exceeds 30% of the local average energy, it is marked as a potential action boundary point. Simultaneously, the degree of change in spectral content is analyzed, using a spectral difference metric. When the spectral difference between adjacent frames exceeds a preset threshold of 0.65, it is also considered a potential boundary point. Combining energy abrupt changes and spectral change points, and applying time constraints (minimum action duration set to 0.3 seconds) for boundary optimization, the precise time interval of each technical action event is finally determined. For example, in a continuous basketball training session, the temporal boundaries of 25 dribbling events (average duration 1.2 seconds), 8 shooting events (average duration 0.8 seconds), and 3 passing events (average duration 0.4 seconds) were successfully detected.

[0029] After determining the time boundaries of technical action events, an acoustic feature subset is extracted for each time interval for technical action type identification. For each identified time interval, a feature subset for that interval is extracted from the original feature set, including time-domain features (such as energy distribution and duration) and frequency-domain features (such as spectral centroid, bandwidth, and energy distribution). A basketball technical action acoustic feature library is pre-established, containing standard acoustic feature representations of various technical actions such as dribbling, shooting, passing, breakthrough, and layup. By calculating the similarity score between the feature subset to be identified and the action type templates in the feature library, a dynamic time warping algorithm is used to process action sequences of different durations, ensuring reasonable alignment between short and long time sequences. The action type with the highest similarity is selected as the identification result, and a minimum confidence threshold of 0.75 is set; identification results below this threshold are marked as "unknown actions." In actual testing, this method achieves an accuracy rate of 95.3% for dribbling, 92.1% for shooting, and 88.7% for passing.

[0030] After determining the action type, the quality of the action is further analyzed. For the identified target technical action type, quality evaluation features associated with the standard execution mode of the action are extracted from the corresponding acoustic feature subset. Taking basketball dribbling as an example, the quality evaluation features include the regularity of dribbling rhythm (quantified by the coefficient of variation of energy peak intervals), the stability of dribbling force (quantified by the standard deviation of peak energy), and the purity of dribbling pitch (quantified by the harmonic noise ratio). A baseline database of acoustic features for various actions performed by athletes of different levels was established, with the baseline data coming from the standard movements collected from professional athletes. For each quality evaluation feature, the normalized deviation between the actual feature value and the baseline feature value is calculated, and a comprehensive deviation score is calculated based on the weight combination of each feature. Finally, the deviation score is converted into a quality score of 0-100 through a mapping function, where 100 points represents perfect execution and 0 points represents serious deviation from the standard. For example, the rhythm regularity deviation of a professional dribbling move is 0.05, the force stability deviation is 0.08, and the tone purity deviation is 0.03. The overall deviation after weighted combination is 0.06, and the final quality score is 94 points, indicating that the dribbling move is close to the professional level.

[0031] After identifying and scoring the technical movements, the target technical movement type and quality score are correlated to generate the final evaluation result, which includes the movement type, start and end times, quality score, and specific improvement suggestions. Different improvement suggestions can be provided for different quality score ranges. For example, for dribbling movements with a score between 60 and 80, the suggestion is "to improve the consistency of dribbling rhythm"; for shooting movements with a score below 60, the suggestion is "insufficient wrist release force, requiring strengthened wrist strength training." The evaluation results can be fed back to the user in real time via a mobile application, and a training report can also be generated, including statistical analysis of various movements and quality evaluation trend charts throughout the training process.

[0032] Through the above technical solution, the present invention realizes intelligent and objective evaluation of basketball technical movements, provides accurate and real-time technical movement feedback for basketball training, and effectively improves training results and technical level.

[0033] In one optional implementation, based on the energy abrupt change points and spectral patterns in the acoustic feature set, the time boundaries of multiple technical action events in basketball are located to obtain the time intervals of each technical action event, including: Time series analysis is performed on the energy distribution characteristics in the acoustic feature set. By calculating the energy change rate between adjacent time windows, time points where the energy change rate exceeds a dynamic threshold are identified as energy mutation points. The dynamic threshold is adaptively adjusted based on the background energy level and energy fluctuation amplitude of the acoustic signal. For each energy mutation point, extract the spectral envelope features within a preset time range before and after the energy mutation point; The spectral envelope features are compared with the pre-stored spectral feature benchmarks for the start and end phases of various basketball techniques. The energy mutation point that matches the spectral feature benchmark for the start phase of the technique is taken as the start boundary point of the technique, and the energy mutation point that matches the spectral feature benchmark for the end phase of the technique is taken as the end boundary point of the technique. Based on the time continuity constraint of basketball technical movements, the starting boundary point and the ending boundary point of the movement that satisfy the continuity constraint in time sequence are paired to obtain the time interval of each technical movement event.

[0034] This implementation first acquires audio signals from basketball games or training sessions, extracts acoustic feature sets after preprocessing, and then locates the time boundaries of multiple technical action events based on energy mutation points and spectral patterns to obtain the time intervals of each technical action event.

[0035] During implementation, sound signals from within the basketball court were acquired via a microphone array at a sampling rate of 48kHz. After noise reduction, the audio signal was segmented into overlapping time windows, each 20 milliseconds in length, with a 50% overlap between adjacent windows. Acoustic feature sets, including time-domain and frequency-domain features, were extracted from each time window. Time-domain features primarily included short-time energy, zero-crossing rate, and energy entropy; frequency-domain features included Mel-frequency cepstral coefficients, spectral centroid, and spectral flux.

[0036] When performing time-series analysis on the energy distribution characteristics in the acoustic feature set, a sliding window method is used to calculate the energy change rate between adjacent time windows. Specifically, for the energy value E(t) at time point t and the energy value E(t-1) at time point t-1, the energy change rate is calculated as |E(t)-E(t-1)| / E(t-1). A sliding window of 2 seconds is maintained to calculate the mean μ and standard deviation σ of the background energy. The dynamic threshold is set to μ+k×σ, where k is an adjustable parameter with an initial value of 2.5. During detection, if no energy abrupt change is detected within 5 consecutive time windows, the k value is automatically reduced by 0.2, with a minimum of 1.5; if more than 10 energy abrupt change points are detected within 0.5 seconds, the k value is increased by 0.3, with a maximum of 3.5. This dynamic adjustment mechanism allows the threshold to adaptively adjust according to the background noise level of different scenarios.

[0037] In practical applications, in a recorded basketball training audio, the background energy had a mean of 0.15 and a standard deviation of 0.05, with an initial dynamic threshold set to 0.275. After a relatively quiet period, the k-value automatically decreased to 2.0, and the threshold was adjusted to 0.25, successfully capturing a slight dribbling sound. In a noisy environment with multiple players training simultaneously, the k-value automatically increased to 3.2, and the threshold increased to 0.31, effectively filtering out interference sounds generated by non-target movements.

[0038] For each identified energy abrupt change point, the spectral envelope features within a 150-millisecond time range before and after that point are extracted. The spectral envelope features are obtained by extracting the envelope line of the spectrum within each time window. Specific steps include: performing a Fast Fourier Transform on the signal within the time window to obtain the spectrum; dividing the spectrum into 25 frequency bands and calculating the average energy value for each band; and connecting the average energy points of each frequency band to form the spectral envelope line. The extracted spectral envelope features form a 25-dimensional feature vector for subsequent matching analysis.

[0039] A pre-established benchmark library of spectral characteristics for basketball techniques is created, containing typical spectral characteristics of the initiation and termination phases of common actions such as shooting, dribbling, passing, stealing, and blocking. Each action in the benchmark library contains at least 50 spectral characteristic samples, obtained from recordings of actions performed by different players under various court conditions. For example, the spectral characteristics of the initiation phase of a shooting motion show lower energy in the low-frequency range (100-500Hz), with a gradual increase in energy in the mid-frequency range (500-2000Hz); while the termination phase exhibits a significant energy peak in the high-frequency range (2000-8000Hz), corresponding to the sound of the ball hitting the net or the backboard.

[0040] During the spectral feature matching process, the cosine similarity algorithm is used to calculate the similarity between the spectral envelope feature to be matched and each action feature in the benchmark library. When the similarity exceeds 0.75, the match is considered successful. In an actual test, the spectral envelope extracted from the beginning of a shooting motion had a similarity of 0.82 with the shooting start feature in the benchmark library, successfully marking it as the shooting start boundary point; the sound of the ball hitting the net at the end of the shooting motion had a similarity of 0.89 with the shooting end feature, and was accurately marked as the end boundary point.

[0041] Based on the time continuity constraints of basketball technical movements, the starting and ending boundary points of movements that satisfy the continuity constraints in time sequence are paired. Reasonable duration ranges for various movements are pre-set; for example, a shooting movement typically lasts 0.8-2.5 seconds, a dribbling movement lasts 0.3-0.8 seconds, and a passing movement lasts 0.5-1.2 seconds. If a shooting starting boundary point is detected and an ending boundary point is found within a time window of 0.8-2.5 seconds, these two boundary points are paired to determine a complete shooting movement, and its time interval is recorded.

[0042] When processing continuous actions, the nearest neighbor principle is used, pairing each starting boundary point with the first satisfying ending boundary point within its subsequent time range. When multiple pairings exist, the pair with the highest similarity product is selected. This method enables accurate location of the time boundaries of various technical actions in complex basketball scenarios, providing a foundation for subsequent action analysis and evaluation.

[0043] In one optional implementation, for each technical action event's time interval, an acoustic feature subset is extracted within the time interval. A matching operation is then performed between the acoustic feature subset and the acoustic feature representations of each basketball technical action type to determine the target technical action type corresponding to the technical action event, including: For each technical action event within a given time interval, the energy distribution within that time interval is extracted from the acoustic feature set. For each basketball technique, acoustic characteristic description parameters are extracted based on the differences in acoustic signals generated at different stages. These parameters are then combined to obtain the acoustic feature representation of the basketball technique. The acoustic feature subset is decomposed into segmented acoustic features corresponding to different stages, and the similarity between each segmented acoustic feature and the acoustic characteristic description parameters of the corresponding stage in the acoustic feature representation of each basketball movement technique is calculated. Weighted fusion of similarities at different stages yields the comprehensive matching degree between the acoustic feature subset and various basketball technique types. The basketball technique type with the highest overall matching degree is selected as the target technique type corresponding to the technique event.

[0044] In practice, the time interval of technical action events can be determined through video image analysis. For example, when a player is detected performing actions such as shooting, dribbling, or passing, the start and end times of the action are recorded to form a time interval. For each defined time interval, a subset of acoustic features within that interval is extracted from the pre-collected sound signals.

[0045] Acoustic feature extraction is fundamental to recognition. For each technical action event within a given time interval, a set of acoustic features for that time period is extracted from pre-recorded audio. These acoustic features include energy distribution, spectral characteristics, and temporal features. Taking energy distribution as an example, by performing short-time analysis on the original audio signal, the energy values ​​of different frequency bands can be calculated. For instance, the frequency range of 20Hz-20kHz can be divided into multiple frequency bands, and the energy value within each band can be calculated. For the sound of a basketball dribbling, its energy is mainly concentrated in the 500Hz-2000Hz frequency band; the sound energy generated when the ball contacts the net during a shot is concentrated in the 1000Hz-3000Hz frequency band; while the energy of a pass is mainly distributed in the 300Hz-1500Hz frequency band.

[0046] Based on the differences in acoustic signals generated at different stages of various basketball techniques, acoustic characteristic parameters can be extracted for each stage. For example, the shooting motion can be divided into three stages: jump, release, and landing, each with unique acoustic characteristics. The jump stage is accompanied by the sound of the shoe sole rubbing against the ground, the release stage has a slight sound of the ball leaving the hand, and the landing stage includes the sound of landing. Similarly, dribbling can be divided into three stages: ball contact with hand, ball descent, and ball contact with the ground, each with different sound characteristics. By collecting a large amount of sample data, acoustic characteristic analysis can be performed on each stage of each technique, ultimately forming an acoustic characteristic representation library.

[0047] In practical applications, taking dribbling as an example, its characteristics can be characterized as follows: during the ball contact phase, the proportion of low-frequency energy is approximately 30%, mid-frequency energy is approximately 50%, and high-frequency energy is approximately 20%; during the ball's descent phase, there are almost no obvious acoustic characteristics; during the ball's ground contact phase, the proportion of low-frequency energy is approximately 15%, mid-frequency energy is approximately 60%, and high-frequency energy is approximately 25%. These values ​​are derived from extensive experimental statistics and can accurately describe the acoustic characteristics of dribbling.

[0048] For each technical action event to be identified, its acoustic feature subset is decomposed into segmented acoustic features corresponding to different stages according to the same stage division criteria. This decomposition process takes into account the temporal nature of the technical action, thus enabling the capture of acoustic changes at each stage of the action. For example, a dribbling action lasting 1.5 seconds is divided into: 0-0.3 seconds for the ball contact stage, 0.3-1.2 seconds for the ball descent stage, and 1.2-1.5 seconds for the ball contact stage. The acoustic features extracted for each stage are then compared with the acoustic features of the corresponding stage of the corresponding technical action in the model library.

[0049] The similarity between the acoustic features of each segment and the corresponding acoustic characteristic description parameters in the acoustic feature representation of each basketball technique type is calculated. Similarity calculation can employ methods such as Euclidean distance and cosine similarity. Taking cosine similarity as an example, the degree of similarity is measured by calculating the cosine of the angle between two feature vectors; the closer the value is to 1, the more similar they are. For instance, comparing the energy distribution features of the ball-touching phase of the dribbling action to be identified [low frequency 18%, mid frequency 58%, high frequency 24%] with the features of the corresponding phase of the dribbling action in the model library [low frequency 15%, mid frequency 60%, high frequency 25%] yields a similarity of 0.998, indicating a high degree of match.

[0050] The importance of acoustic features at different stages for action recognition varies, therefore, a weighted fusion of similarities at different stages is necessary. The weight values ​​can be optimized and determined through extensive experimental data. Taking dribbling as an example, the weight for the ball-touching stage can be set to 0.3, the ball-falling stage to 0.1, and the ball-hitting stage to 0.6. The comprehensive matching degree between the acoustic feature subset and each basketball technique type is calculated using a weighted summation method.

[0051] Finally, the basketball technique type with the highest overall matching degree is selected as the target technique type corresponding to the technique event. For example, if a subset of acoustic features has an overall matching degree of 0.92 with dribbling, 0.75 with shooting, and 0.68 with passing, then the technique event will be identified as a dribbling action.

[0052] Through the above, this invention can accurately identify various technical movements in basketball games, providing technical support for applications such as basketball training analysis and game replay annotation. Experimental results show that the method can achieve an accuracy rate of over 90% in recognizing common basketball technical movements (such as dribbling, passing, shooting, layups, and dunks), demonstrating high practical value.

[0053] In one optional implementation, for each type of basketball technique, based on the differences in acoustic signals generated at different stages, acoustic characteristic description parameters for each stage are extracted, and the acoustic characteristic description parameters for each stage are combined to obtain the acoustic feature representation of the basketball technique type, including: For each type of basketball technique, based on the physical execution process of that technique, it is divided into multiple consecutive execution stages. For each execution stage of each basketball technique, multiple sets of standard execution acoustic samples are collected for that execution stage. These standard execution acoustic samples are derived from the acoustic signal records generated at each execution stage of that basketball technique under standard execution conditions. For each execution stage, a time-frequency domain analysis is performed on the standard execution acoustic samples to extract acoustic characteristic descriptive parameters that characterize the distribution of acoustic signals in the energy domain, frequency domain, and time domain of that execution stage. The acoustic characteristic description parameters extracted from multiple sets of standard execution acoustic samples in the same execution phase are statistically aggregated to obtain representative acoustic characteristic description parameters for that execution phase. The representative acoustic characteristic parameters of each execution stage of the basketball technique are combined according to the time sequence of the execution stages to obtain the acoustic feature representation of the basketball technique.

[0054] Figure 2 A schematic flowchart illustrating the acoustic feature representation of determining basketball technique types according to an embodiment of the present invention. Figure 2 As shown, each basketball technique can be pre-divided into multiple consecutive execution stages, and the features of each stage can be extracted based on the differences in acoustic signals. Taking the shooting action as an example, it can be divided into four consecutive execution stages: ball preparation, arm raising, release, and landing.

[0055] For the ball-handling preparation phase of a shooting motion, 10 sets of standard acoustic samples were collected. Each set of samples was performed by a professional basketball player in a standard basketball court and recorded using a high-precision microphone array (sampling rate 48kHz, 24-bit quantization). Time-frequency domain analysis was performed on the collected acoustic samples, extracting energy domain features including short-time mean energy of 23.5dB, energy variance of 5.2dB, and peak energy of 25.7dB; frequency domain features including spectral centroid of 872Hz, spectral entropy of 0.76, and frequency band energy ratio (38% in the low-frequency range of 0-500Hz, 42% in the mid-frequency range of 500-2000Hz, and 20% in the high-frequency range of 2000-8000Hz); and time domain features including zero-crossing rate of 328 times / second and autocorrelation coefficient of 0.65. By statistically aggregating the 10 sets of samples, the mean and standard deviation of each feature parameter were calculated to obtain representative acoustic characteristic descriptive parameters for the ball-handling preparation phase.

[0056] For the boom-raising phase, 10 sets of standard acoustic samples were collected, and acoustic features were extracted. The energy domain characteristics of this phase showed that the short-time mean energy increased to 27.8 dB, the energy variance decreased to 3.1 dB, and the peak energy reached 30.2 dB. The frequency domain characteristics showed that the spectral centroid increased to 1245 Hz, the spectral entropy decreased to 0.65, and the frequency band energy ratio changed to 25% low frequency, 55% mid frequency, and 20% high frequency. The time domain characteristics showed that the zero-crossing rate increased to 452 times / second, and the autocorrelation coefficient decreased to 0.53. After statistical aggregation, representative acoustic characteristic descriptive parameters for the boom-raising phase were obtained.

[0057] The release phase is the crucial stage of the shooting motion, during which the acoustic characteristics exhibit significant changes. Analysis of 10 standard samples shows that during this phase, the short-time mean energy reaches a peak of 32.5 dB, the energy variance is 8.7 dB, and the peak energy is 38.6 dB. In the frequency domain, the spectral centroid rapidly rises to 2356 Hz, the spectral entropy drops to 0.48, and the frequency band energy ratio shifts to 15% low frequency, 35% mid frequency, and 50% high frequency. In the time domain, the zero-crossing rate reaches as high as 867 times / second, and the autocorrelation coefficient drops to 0.32. The statistically aggregated representative parameters accurately reflect the characteristic acoustic signal generated when the ball leaves the hand.

[0058] The acoustic characteristics during the landing phase reflect the state after the shot. Analysis of 10 samples shows that during this phase, the short-term mean energy rapidly decreases to 18.3 dB, the energy variance increases to 9.5 dB, and the peak energy reaches 36.2 dB when the ball touches the net or backboard. In the frequency domain, the spectral centroid drops to 1025 Hz, the spectral entropy increases to 0.82, and the frequency band energy ratio becomes 45% low-frequency, 35% mid-frequency, and 20% high-frequency. In the time domain, the zero-crossing rate decreases to 352 times / second, and the autocorrelation coefficient increases to 0.68. The representative parameters obtained after statistical aggregation fully characterize the acoustic properties during the landing phase.

[0059] The representative acoustic characteristic description parameters of the four execution stages of the shooting motion are combined in chronological order to form a complete acoustic feature representation of the shooting motion. This representation is stored in the form of a feature vector, containing 32 acoustic characteristic description parameters for the four stages, forming a 128-dimensional feature vector.

[0060] Similarly, the dribbling action can be divided into four execution stages: ball contact with the hand, ball descent, ball impact, and ball rebound. Ten sets of standard acoustic samples were collected for each stage, and time-frequency analysis was performed. The acoustic characteristics of the ball impact stage are particularly significant. In the energy domain, the short-time mean energy is 35.7 dB, the energy variance is 4.2 dB, and the peak energy is 40.3 dB. In the frequency domain, the spectral centroid is 1856 Hz, the spectral entropy is 0.51, and the frequency band energy ratio is 20% low frequency, 30% mid frequency, and 50% high frequency. In the time domain, the zero-crossing rate is 783 times / second, and the autocorrelation coefficient is 0.38. After statistical aggregation and combination of representative parameters from the four stages, a complete acoustic characterization of the dribbling action is obtained.

[0061] The passing motion is divided into four execution phases: ball preparation, release, ball flight, and receiving. The acoustic characteristics of the passing release phase differ significantly from those of the shooting release phase. In the energy domain, the short-time mean energy is 28.6 dB, the energy variance is 6.3 dB, and the peak energy is 33.8 dB. In the frequency domain, the spectral centroid is 1678 Hz, the spectral entropy is 0.57, and the frequency band energy ratio is 28% low-frequency, 45% mid-frequency, and 27% high-frequency. In the time domain, the zero-crossing rate is 625 times / second, and the autocorrelation coefficient is 0.45. The acoustic characteristics of the passing motion are obtained by combining representative parameters from these four phases.

[0062] Based on the above, a complete acoustic feature representation library can be established for various technical movements in basketball (such as shooting, dribbling, passing, layup, dunking, defensive slide, etc.), providing basic data support for subsequent basketball action recognition and evaluation based on acoustic signals.

[0063] In one optional implementation, for the target technical action type, quality evaluation features associated with the standard execution mode of the target technical action type are extracted from the acoustic feature subset, and a quality score for the technical action event is determined based on the deviation between the quality evaluation features and a preset standard acoustic feature benchmark for the target technical action type, including: Identify multiple quality evaluation dimensions for assessing the execution quality of the target technical action type; For each quality evaluation dimension, extract the quality evaluation features associated with that quality evaluation dimension from the subset of acoustic features; For each quality evaluation dimension, a preset standard acoustic feature benchmark for the target technical action type on that quality evaluation dimension is obtained, and the deviation between the quality evaluation feature corresponding to that quality evaluation dimension and the standard acoustic feature benchmark is calculated. Based on the deviation amount of each quality evaluation dimension, a deviation vector is determined. The deviation vector is then matched with the deviation vector templates corresponding to multiple quality levels of the target technical action type in a pre-stored manner to identify the quality level with the highest similarity to the deviation vector. The quality score of the technical action event is determined based on the similarity between the deviation vector and the deviation vector template corresponding to the quality level, and the difference in similarity between the deviation vector and the deviation vector templates corresponding to adjacent quality levels.

[0064] In the implementation process, it is first necessary to determine multiple quality evaluation dimensions for assessing the execution quality of the target technical movement type. Taking basketball shooting technique as an example, quality evaluation dimensions such as force stability, rhythm continuity, and wrist release smoothness can be set. These dimensions are selected based on the correlation between acoustic characteristics and movement quality, and are jointly determined by professional coaches and acoustic analysis experts.

[0065] For each quality evaluation dimension, relevant quality evaluation features are extracted from the acoustic feature subset. For example, for the force stability dimension, the smoothness and rate of change of the sound energy curve are extracted; for the rhythm continuity dimension, the regularity index of the time interval between sound peaks is extracted; and for the wrist release smoothness dimension, the proportion of high-frequency components and frequency distribution characteristics of the sound spectrum are extracted. Specifically, taking the shooting action as an example, the sound signal during the shooting process can be recorded, and the distribution changes of sound energy in different frequency bands can be extracted through time-frequency analysis. When the energy distribution of the sound in the 50-200Hz frequency band shows a characteristic of first slowly rising and then rapidly falling, and the high-frequency components (800-1200Hz) have a significant peak at the release point, it indicates that the wrist release action is smooth.

[0066] For each quality evaluation dimension, a standard acoustic feature benchmark for the target technical action type in that dimension is obtained, and the deviation between the quality evaluation feature and the standard benchmark is calculated. Taking basketball shooting as an example, the standard acoustic feature benchmark can be derived from the statistical results of acoustic feature data generated by collecting acoustic feature data from multiple professional athletes performing standard shooting actions. For example, during the power generation phase of a standard shooting action, the energy curve in the 200-500Hz frequency band should show a steady rise lasting 0.3±0.05 seconds, and the deviation is calculated by the difference between the actual collected sound features and this standard pattern.

[0067] Based on the deviations of each quality evaluation dimension, a deviation vector is constructed, which can be represented as (force stability deviation value, rhythm continuity deviation value, wrist release smoothness deviation value). For example, the deviation vector of a certain shooting motion is (0.15, 0.08, 0.22), indicating the degree of deviation from the standard pattern in these three dimensions.

[0068] The aforementioned deviation vector is matched with pre-stored deviation vector templates corresponding to multiple quality levels of the target technical action type. For example, a shooting action can be preset with five quality levels: excellent, good, average, passable, and failable, each level corresponding to a preset deviation vector template. The template for excellent level is (0.05, 0.03, 0.04), and the template for good level is (0.10, 0.08, 0.12), etc. By calculating the Euclidean distance or cosine similarity between the actual deviation vector and each template, the quality level with the highest similarity to the actual deviation vector is identified.

[0069] After determining the best-matching quality level, the similarity difference between the deviation vector and the deviation vector templates corresponding to adjacent quality levels is further considered to achieve a more refined score. Specifically, if the deviation vector (0.15, 0.08, 0.22) has the highest similarity to the "Medium" level template (0.20, 0.15, 0.25), a small difference in similarity to the "Good" level template (0.10, 0.08, 0.12), and a large difference in similarity to the "Pass" level template (0.30, 0.25, 0.35), the final score can be adjusted to above-average.

[0070] The quality score can be calculated using a weighted average method. Assuming a base score of 70 for the "Medium" grade, a similarity of 0.85 with the "Medium" grade template, 0.75 with the "Good" grade template, and 0.60 with the "Pass" grade template, the score can be weighted according to the similarity difference. The specific calculation is: 70 + (0.85-0.60) / (0.85-0.60)×(0.75-0.85) / (0.75-0.60)×10 = 74 points. This indicates that the quality score for this technical movement is 74 points, which is above average.

[0071] Through the above methods, the execution quality of technical movements can be objectively and meticulously evaluated based on acoustic characteristics, providing a scientific basis for technical movement training. Furthermore, the scoring criteria can be adaptively adjusted according to the different levels of users, providing targeted quality assessment feedback for beginners and professionals alike.

[0072] In one optional implementation, for each quality evaluation dimension, a preset standard acoustic feature benchmark for the target technical action type on that quality evaluation dimension is obtained, and the deviation between the quality evaluation feature corresponding to that quality evaluation dimension and the standard acoustic feature benchmark is calculated, including: For the target technical action type, acoustic samples of the target technical action type at multiple different execution quality levels are collected in advance, and each acoustic sample is labeled with its corresponding execution quality level; The acoustic samples are grouped into a standard sample set according to the performance quality level. For each quality evaluation dimension, acoustic samples with a preset excellent performance quality level are selected from the standard sample set as benchmark samples. Acoustic feature analysis is performed on the benchmark sample, and the acoustic feature statistics corresponding to the quality evaluation dimension are extracted as the standard acoustic feature benchmark for the quality evaluation dimension. The initial deviation between the quality evaluation feature corresponding to the quality evaluation dimension and the standard acoustic feature benchmark is calculated. Based on the positional relationship between the quality evaluation features corresponding to the quality evaluation dimension and the acoustic feature distribution of acoustic samples with different performance quality levels in the standard sample set on the quality evaluation dimension, the deviation correction coefficient of the quality evaluation dimension is determined. The initial deviation of the quality evaluation dimension is calculated by combining it with the deviation correction coefficient to obtain the deviation of the quality evaluation dimension.

[0073] In practical applications, evaluation benchmarks need to be established for specific target technical movement types. Taking the basketball "standard shot" as an example, this movement produces specific acoustic characteristics during execution, such as the friction sound between the ball and fingers upon release, the air sound of the ball spinning after the shot, the sound of the ball hitting the net, and the sound of the backboard bouncing. An acoustic sample library can be constructed by collecting acoustic signals generated when professional players and practitioners of different skill levels perform this movement. The acquisition equipment includes a high-precision microphone array with a sampling rate set to 48kHz to ensure the capture of minute acoustic changes during the shooting process.

[0074] During the data collection process, 10 professional basketball coaches were invited to score each shooting motion on-site. The scoring dimensions included: release smoothness, ball spin stability, shooting arc, and power control, with each dimension quantified on a scale of 1-10. Based on the scoring results, the acoustic samples were divided into four quality levels: Excellent (8-10 points), Good (6-8 points), Average (4-6 points), and Poor (below 4 points). For each quality evaluation dimension, 20 samples were randomly selected from the Excellent group as the baseline sample set for that dimension.

[0075] Acoustic features were extracted and analyzed from the benchmark samples. Taking the "ball rotation stability" dimension as an example, the following statistical features were extracted through acoustic processing: the spectral characteristics of the friction between the ball and fingers at the moment of release (average frequency range of 2000-3500Hz), the frequency characteristics of the airflow sound generated by the ball's rotation in the air (main frequency stable between 850-950Hz), and the duration of the rotation sound (average value of 0.65 seconds). These statistical measures were used as the standard acoustic feature benchmark for the "ball rotation stability" dimension.

[0076] When evaluating new shooting execution samples, the same acoustic features are extracted, and the initial deviation from the baseline is calculated. For example, when a trainee performs a "standard shot," the frequency range of the release friction sound is detected to be 1800-3000Hz, which is narrower and lower than the baseline value; the dominant frequency of the ball rotation airflow sound is 780Hz, lower than the baseline range, indicating a frequency deviation; the duration of the rotation sound is 0.58 seconds, 0.07 seconds shorter than the baseline value. These raw deviation data are recorded as the initial deviation.

[0077] To more accurately reflect performance quality, deviation correction coefficients need to be determined. The acoustic characteristic distribution of different quality level groups in the standard sample set is analyzed to establish a mapping relationship between deviation and quality score. Taking the "ball rotation stability" dimension as an example, analysis reveals that: for every 500Hz decrease in the spectral range of the hand friction sound, the quality score decreases by an average of 1.8 points; for every 100Hz deviation of the dominant frequency of the rotating airflow sound from the baseline range, the quality score decreases by an average of 2.2 points; and for every 0.1 seconds decrease in the duration of the rotating sound, the quality score decreases by an average of 1.5 points. Based on these analytical results, a deviation correction coefficient matrix for the "ball rotation stability" dimension is determined.

[0078] For the example above, the corrected deviations are calculated as follows: a friction sound frequency deviation of approximately 700 Hz corresponds to a correction factor of 0.65, resulting in a deviation of -2.5 points; a rotating airflow sound frequency deviation of -70 Hz corresponds to a correction factor of 0.75, resulting in a deviation of -1.16 points; and a duration deviation of -0.07 seconds corresponds to a correction factor of 0.9, resulting in a deviation of -0.95 points. Combining these three corrected deviations, the overall deviation for the trainee in the "ball rotation stability" dimension is calculated to be -4.61 points.

[0079] The same process is applied to other quality evaluation dimensions. For the "smoothness of release" dimension, the sound features produced when the fingers release the ball are extracted, such as the continuity and smoothness of the sound. For the "shooting arc" dimension, the frequency changes and duration patterns of the air sound during the ball's flight are analyzed. For the "power control" dimension, the characteristics of the release sound intensity and the sound of the ball hitting the net or backboard are evaluated.

[0080] Regarding the "power control" dimension, excellent samples exhibit moderate release sound intensity (average 65dB), a crisp and concentrated net touch sound (frequency concentrated between 300-500Hz), and minimal or no backboard bounce sound. Power control-related indicators were extracted from the acoustic characteristics, including release sound energy distribution, net touch sound spectrum characteristics, and backboard bounce sound intensity. When a practitioner's release sound intensity was detected to be excessively high (75dB), the net touch sound energy was dispersed, and the backboard bounce sound was prominent, it was determined that they had a significant deviation in the power control dimension.

[0081] In practical applications, different tolerance levels can be set for different user groups. For example, for young basketball enthusiasts, a more lenient tolerance range can be set, allowing rotational sound frequency deviations within ±150Hz; while for professional players, a stricter standard can be set, requiring rotational sound frequency deviations to be controlled within ±50Hz. The deviation correction coefficient can be dynamically adjusted based on the user's progress, providing more personalized quality evaluation results.

[0082] Acoustic assessment of the shooting motion can provide specific suggestions for improvement. For example, if the ball spin stability deviation is mainly due to the release friction sound characteristics, it is recommended that the trainee improve the contact method between the fingers and the ball and the release technique; if the acoustic characteristics related to the shooting arc are found to be low, it is recommended to increase the shooting angle and power.

[0083] The above methods enable objective quantification of the execution quality of basketball techniques, providing users with precise feedback and improvement suggestions. This method is not only applicable to shooting evaluation but can also be extended to the quality assessment of various basketball techniques such as dribbling, passing, and breakthroughs, demonstrating broad application prospects.

[0084] In one optional implementation, the target technical movement type and the quality score are correlated to generate an evaluation result for the basketball technical movement, including: Based on the score range in which the quality score falls, the quality score is mapped to a corresponding quality level identifier, wherein the quality level identifier represents the execution quality level of the technical action event under the target technical action type; Based on the target technical action type and the deviation of each quality evaluation dimension, the quality evaluation dimensions with deviations exceeding the preset deviation threshold are identified as dimensions to be improved. For each dimension to be improved, the correlation between the dimension to be improved and the execution technical elements of the target technical action type is obtained, and improvement suggestion information for the corresponding execution technical elements is generated based on the correlation. The target technical movement type, the quality score, the quality level identifier, and the improvement suggestion information are linked and organized to generate an evaluation result for basketball technical movements.

[0085] The evaluation results of basketball technical movements are generated by associating the target technical movement type with the quality score. The specific implementation method is as follows: In this embodiment, the process of mapping quality scores to quality level identifiers includes pre-setting multiple score ranges and their corresponding quality level identifiers. For example, 0-60 points can be set as "Fail", 60-70 points as "Pass", 70-80 points as "Good", 80-90 points as "Excellent", and 90-100 points as "Outstanding". When the quality score for a target technical action type such as "shooting" is obtained as 85 points, it will be automatically mapped to the "Excellent" quality level identifier.

[0086] When identifying dimensions for improvement, the deviation of each quality evaluation dimension is analyzed. Taking shooting motion as an example, the quality evaluation dimensions involved include release angle, wrist strength control, body balance, and release timing. The deviation of each dimension is calculated; for example, a release angle deviation of 5 degrees, a wrist strength control deviation of 10%, a body balance deviation of 3%, and a release timing deviation of 15%. Assuming the preset deviation thresholds are 3 degrees for release angle, 8% for wrist strength control, 5% for body balance, and 10% for release timing, then release angle and release timing are identified as dimensions for improvement.

[0087] For identified dimensions requiring improvement, the correlation between them and the execution technical elements of the target technical action type is obtained. This involves a pre-stored database of correlations between various technical action types and quality evaluation dimensions. For example, for the release angle dimension of a shooting motion, the associated execution technical elements include elbow position, wrist flexion, and release point height. For the release timing dimension, the associated execution technical elements include jump timing and the moment of ball release.

[0088] Based on the acquired correlations, targeted improvement suggestions are generated. For the release angle dimension, the improvement suggestion is: "Your release angle is 5 degrees too low. It is recommended to adjust the elbow position to the same height as the shoulder, increase the wrist flexion by about 10 degrees, and keep the release point slightly above and to the right of the head." For the release timing dimension, the improvement suggestion is: "Your release timing is 15% too late. It is recommended to release the ball before reaching the highest point of the jump, rather than during the descent, to ensure better stability and accuracy."

[0089] By linking and organizing the target technique type, quality score, quality level label, and improvement suggestions, a complete evaluation result will be generated, such as: "Technique type: Shooting; Quality score: 85 points; Quality level label: Excellent; Improvement suggestions: 1. Regarding the release angle, your release angle is 5 degrees too low. It is recommended to adjust the elbow position to the same height as the shoulder, increase the wrist flexion by about 10 degrees, and keep the release point slightly above and to the right of the head; 2. Regarding the release timing, your release timing is 15% too late. It is recommended to release the ball before reaching the highest point of the jump, rather than during the descent, to ensure better stability and accuracy." In another embodiment, the evaluation can be further refined for different types of shooting motions. For example, for three-point shots, more attention is paid to power control and arc requirements. When an athlete is detected performing a three-point shooting motion, the quality score is 76 points, which is mapped to a "good" quality level. By analyzing the deviations in various dimensions, it was found that the shooting arc deviation was 12 degrees (the preset threshold is 8 degrees), and the elbow abduction angle deviation was 7 degrees (the preset threshold is 5 degrees). The generated improvement suggestions could include: "It is recommended to increase the shooting arc by approximately 12 degrees, which can be achieved by raising the release point and increasing the upward flexion force of the wrist; at the same time, pay attention to keeping the elbow pointing towards the basket and reducing the abduction angle by 7 degrees, which helps to improve shooting stability." For dribbling techniques, the quality evaluation dimensions include finger strength control, body center of gravity shifts, and field of vision. Assuming an athlete's dribbling quality score is 92, it corresponds to an "Excellent" quality level. Analysis shows that only the field of vision deviation exceeds the threshold by 25% (preset threshold is 20%). The generated improvement suggestion is: "Your overall dribbling technique is excellent, but there is still room for improvement in field of vision. It is recommended to raise your head while dribbling, increase the overall field of vision, and perform a full-court scan before each change of direction. This will help improve decision-making and passing opportunities during the game." This method can also be applied to the evaluation of other technical movements in basketball, such as defense, passing, and rebounding. By analyzing specific quality evaluation dimensions for each technical movement type, targeted suggestions for improvement can be provided to help athletes train in a targeted manner and improve their basketball skills.

[0090] The intelligent evaluation system for basketball technique movements based on acoustic features according to embodiments of the present invention includes: The first unit is used to acquire acoustic signals generated during basketball movements, perform time-frequency domain decomposition on the acoustic signals, and extract a set of acoustic features. The second unit is used to locate the time boundaries of multiple technical action events in basketball based on the energy mutation points and spectral patterns in the acoustic feature set, and to obtain the time interval of each technical action event. The third unit is used to extract a subset of acoustic features within the time interval of each technical action event, and perform matching operations between the subset of acoustic features and the acoustic feature representations of each basketball technical action type to determine the target technical action type corresponding to the technical action event. The fourth unit is used to extract quality evaluation features associated with the standard execution mode of the target technical action type from the acoustic feature subset for the target technical action type, and to determine the quality score of the technical action event based on the deviation between the quality evaluation features and the preset standard acoustic feature benchmark of the target technical action type. The fifth unit is used to associate the target technical action type with the quality score to generate an evaluation result for basketball technical actions.

[0091] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0092] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0093] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent evaluation of basketball technical movements based on acoustic features, characterized in that, include: Acoustic signals generated during basketball play are acquired, and the acoustic signals are decomposed in the time-frequency domain to extract a set of acoustic features. Based on the energy mutation points and spectral patterns in the acoustic feature set, the time boundaries of multiple technical action events in basketball are located to obtain the time interval of each technical action event; For each technical action event within a given time interval, an acoustic feature subset is extracted. Based on this subset, a matching operation is performed between the acoustic feature subset and the acoustic feature representations of each basketball technical action type to determine the target technical action type corresponding to the technical action event. For the target technical action type, quality evaluation features associated with the standard execution mode of the target technical action type are extracted from the acoustic feature subset, and a quality score for the technical action event is determined based on the deviation between the quality evaluation features and a preset standard acoustic feature benchmark for the target technical action type. The target technical movement type and the quality score are correlated to generate an evaluation result for basketball technical movements.

2. The method according to claim 1, characterized in that, Based on the energy abrupt change points and spectral patterns in the acoustic feature set, the temporal boundaries of multiple technical action events in basketball are located to obtain the time intervals of each technical action event, including: Time series analysis is performed on the energy distribution characteristics in the acoustic feature set. By calculating the energy change rate between adjacent time windows, time points where the energy change rate exceeds a dynamic threshold are identified as energy mutation points. The dynamic threshold is adaptively adjusted based on the background energy level and energy fluctuation amplitude of the acoustic signal. For each energy mutation point, extract the spectral envelope features within a preset time range before and after the energy mutation point; The spectral envelope features are compared with the pre-stored spectral feature benchmarks for the start and end phases of various basketball techniques. The energy mutation point that matches the spectral feature benchmark for the start phase of the technique is taken as the start boundary point of the technique, and the energy mutation point that matches the spectral feature benchmark for the end phase of the technique is taken as the end boundary point of the technique. Based on the time continuity constraint of basketball technical movements, the starting boundary point and the ending boundary point of the movement that satisfy the continuity constraint in time sequence are paired to obtain the time interval of each technical movement event.

3. The method according to claim 1, characterized in that, For each technical action event within a given time interval, an acoustic feature subset is extracted. Based on this subset, a matching operation is performed between the acoustic feature subset and the acoustic feature representations of various basketball technical action types to determine the target technical action type corresponding to the technical action event, including: For each technical action event within a given time interval, the energy distribution within that time interval is extracted from the acoustic feature set. For each basketball technique, acoustic characteristic description parameters are extracted based on the differences in acoustic signals generated at different stages. These parameters are then combined to obtain the acoustic feature representation of the basketball technique. The acoustic feature subset is decomposed into segmented acoustic features corresponding to different stages, and the similarity between each segmented acoustic feature and the acoustic characteristic description parameters of the corresponding stage in the acoustic feature representation of each basketball movement technique is calculated. Weighted fusion of similarities at different stages yields the comprehensive matching degree between the acoustic feature subset and various basketball technique types. The basketball technique type with the highest overall matching degree is selected as the target technique type corresponding to the technique event.

4. The method according to claim 3, characterized in that, For each basketball technique, based on the differences in acoustic signals generated at different stages, acoustic characteristic description parameters for each stage are extracted. These parameters are then combined to obtain the acoustic feature representation of the basketball technique, including: For each type of basketball technique, based on the physical execution process of that technique, it is divided into multiple consecutive execution stages. For each execution stage of each basketball technique, multiple sets of standard execution acoustic samples are collected for that execution stage. These standard execution acoustic samples are derived from the acoustic signal records generated at each execution stage of that basketball technique under standard execution conditions. For each execution stage, a time-frequency domain analysis is performed on the standard execution acoustic samples to extract acoustic characteristic descriptive parameters that characterize the distribution of acoustic signals in the energy domain, frequency domain, and time domain of that execution stage. The acoustic characteristic description parameters extracted from multiple sets of standard execution acoustic samples in the same execution phase are statistically aggregated to obtain representative acoustic characteristic description parameters for that execution phase. The representative acoustic characteristic parameters of each execution stage of the basketball technique are combined according to the time sequence of the execution stages to obtain the acoustic feature representation of the basketball technique.

5. The method according to claim 1, characterized in that, For the target technical action type, quality evaluation features associated with the standard execution mode of the target technical action type are extracted from the acoustic feature subset. Based on the deviation between the quality evaluation features and a preset standard acoustic feature benchmark for the target technical action type, a quality score for the technical action event is determined, including: Identify multiple quality evaluation dimensions for assessing the execution quality of the target technical action type; For each quality evaluation dimension, extract the quality evaluation features associated with that quality evaluation dimension from the subset of acoustic features; For each quality evaluation dimension, a preset standard acoustic feature benchmark for the target technical action type on that quality evaluation dimension is obtained, and the deviation between the quality evaluation feature corresponding to that quality evaluation dimension and the standard acoustic feature benchmark is calculated. Based on the deviation amount of each quality evaluation dimension, a deviation vector is determined. The deviation vector is then matched with the deviation vector templates corresponding to multiple quality levels of the target technical action type in a pre-stored manner to identify the quality level with the highest similarity to the deviation vector. The quality score of the technical action event is determined based on the similarity between the deviation vector and the deviation vector template corresponding to the quality level, and the difference in similarity between the deviation vector and the deviation vector templates corresponding to adjacent quality levels.

6. The method according to claim 5, characterized in that, For each quality evaluation dimension, a preset standard acoustic feature benchmark for the target technical action type on that quality evaluation dimension is obtained, and the deviation between the quality evaluation feature corresponding to that quality evaluation dimension and the standard acoustic feature benchmark is calculated, including: For the target technical action type, acoustic samples of the target technical action type at multiple different execution quality levels are collected in advance, and each acoustic sample is labeled with its corresponding execution quality level; The acoustic samples are grouped into a standard sample set according to the performance quality level. For each quality evaluation dimension, acoustic samples with a preset excellent performance quality level are selected from the standard sample set as benchmark samples. Acoustic feature analysis is performed on the benchmark sample, and the acoustic feature statistics corresponding to the quality evaluation dimension are extracted as the standard acoustic feature benchmark for the quality evaluation dimension. The initial deviation between the quality evaluation feature corresponding to the quality evaluation dimension and the standard acoustic feature benchmark is calculated. Based on the positional relationship between the quality evaluation features corresponding to the quality evaluation dimension and the acoustic feature distribution of acoustic samples with different performance quality levels in the standard sample set on the quality evaluation dimension, the deviation correction coefficient of the quality evaluation dimension is determined. The initial deviation of the quality evaluation dimension is calculated by combining it with the deviation correction coefficient to obtain the deviation of the quality evaluation dimension.

7. The method according to claim 1, characterized in that, The target technical movement type and the quality score are correlated to generate an evaluation result for basketball technical movements, including: Based on the score range in which the quality score falls, the quality score is mapped to a corresponding quality level identifier, wherein the quality level identifier represents the execution quality level of the technical action event under the target technical action type; Based on the target technical action type and the deviation of each quality evaluation dimension, the quality evaluation dimensions with deviations exceeding the preset deviation threshold are identified as dimensions to be improved. For each dimension to be improved, the correlation between the dimension to be improved and the execution technical elements of the target technical action type is obtained, and improvement suggestion information for the corresponding execution technical elements is generated based on the correlation. The target technical movement type, the quality score, the quality level identifier, and the improvement suggestion information are linked and organized to generate an evaluation result for basketball technical movements.

8. An intelligent evaluation system for basketball technical movements based on acoustic features, used to implement the method as described in any one of claims 1-7, characterized in that, include: The first unit is used to acquire acoustic signals generated during basketball movements, perform time-frequency domain decomposition on the acoustic signals, and extract a set of acoustic features. The second unit is used to locate the time boundaries of multiple technical action events in basketball based on the energy mutation points and spectral patterns in the acoustic feature set, and to obtain the time interval of each technical action event. The third unit is used to extract a subset of acoustic features within the time interval of each technical action event, and perform matching operations between the subset of acoustic features and the acoustic feature representations of each basketball technical action type to determine the target technical action type corresponding to the technical action event. The fourth unit is used to extract quality evaluation features associated with the standard execution mode of the target technical action type from the acoustic feature subset for the target technical action type, and to determine the quality score of the technical action event based on the deviation between the quality evaluation features and the preset standard acoustic feature benchmark of the target technical action type. The fifth unit is used to associate the target technical action type with the quality score to generate an evaluation result for basketball technical actions.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images