Quantitative analysis and feedback method for rhythm synchronism of sports dance training

Abstract

The invention discloses a quantitative analysis and feedback method for rhythm synchronism of sports dance training, and relates to the technical field of data analysis of sports dance training, comprising the following steps: training data multi-modal acquisition, acquisition of 16 key joint data by 12 cameras, acquisition of audio by 48kHz directional microphone wind, and acquisition of heart rate by a wrist cardiotachometer; preprocessing, motion data noise filtering, complementing and abnormity removing, audio feature extraction and noise removing, heart rate smoothing and time sequence unifying are carried out; bidirectional extraction of rhythm features: identifying a rhythm period and accent as well as action lifting force and an action period by audio; synchronism quantification is carried out, time difference and sequence consistency are calculated, and a synchronism decline segment is identified; analyzing results, grading, marking weak links, and calculating progress amplitude; and feedback, AR and voice prompt are generated, and a plan pushing APP is called. The method solves the problem of subjective fuzzy judgment of traditional training synchronism; real-time multi-sensory feedback helps real-time adjustment, and personalized reports and plans meet requirements.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sports dance training data analysis, and particularly relates to a quantitative analysis and feedback method for rhythm synchronicity in sports dance training. BACKGROUND

[0002] In sports dance training, rhythm synchronicity is a core indicator for measuring training quality, and directly affects the fluency and watchability of dance performance. In the traditional training mode, the judgment of rhythm synchronicity highly depends on the subjective experience of the coach. The coach can only give vague evaluations such as "action is slightly fast" and "rhythm is not followed" by observing the matching situation of dancer's action and music with naked eyes, and cannot quantify the time difference between action and music beat, the consistency of action sequence and beat sequence, and other key dimensions, which leads to the difficulty of dancers to accurately locate the problem and low training efficiency. For example, the deviation of dancer's hip joint force moment and music accent may reach 120 ms, but the coach can only perceive that "the force moment is not right", and cannot specify the specific value of the deviation, and the dancer still has difficulty in achieving ideal synchronization state after repeated adjustment.

[0003] The existing training auxiliary means have the problems of single data collection and limited analysis dimensions. Some auxiliary tools only collect action data through a camera, or only analyze music beat through an audio device, without realizing the fusion and linkage of multi-modal data, ignoring the correlation between training intensity and rhythm adaptation state reflected by the dancer's heart rate, and not considering the synchronicity of muscle force details and music rhythm. When the dancer appears fatigue due to too high training intensity (heart rate exceeding 150 times per minute), the action synchronicity will decrease significantly, but the existing tools cannot correlate and analyze the fatigue factor and synchronicity deviation, and only attribute it to "insufficient rhythm perception", which leads to poor pertinence of improvement suggestions. At the same time, for different styles of sports dance such as waltz and tango, their beat characteristics (such as 3 / 4 beat and 4 / 4 beat, and cut rhythm) have significant differences, but the existing tools mostly adopt a unified quantification standard, which cannot adapt to the rhythm requirements of different styles, and the analysis accuracy is limited.

[0004] The synchronization analysis in the multi-dancer cooperative training scene is a short board of the prior art. In double or multi-person sports dancing (such as double waltz and group Latin), the motion synchronization between dancers is as important as the synchronization of individual dancers and music, but the existing tools can only evaluate the synchronization of each dancer with the music, cannot calculate the time difference of the same action between dancers, the intra-class similarity of the action sequence, and cannot mark the weak dancer combination. For example, in double training, the time difference of the turning action of dancer A and dancer B reaches 60 ms, which cannot be effectively identified, resulting in that the cooperative training is long-term in the "rely on tacit understanding" stage, and it is difficult to improve the cooperative efficiency through accurate analysis. In addition, the existing feedback mode is mainly to generate a simple report after training, which lacks real-time interactive feedback, and dancers cannot adjust the action in time during training, and the improvement suggestions in the report are mostly general templates, which are not combined with historical training data of dancers to generate personalized solutions, and it is difficult to support long-term training improvement. SUMMARY

[0005] The present application provides a quantitative analysis and feedback method for rhythm synchronization in sports dancing training to solve the problems mentioned in the prior art.

[0006] To achieve the above purpose, the present application adopts the following technical scheme: a quantitative analysis and feedback method for rhythm synchronization in sports dancing training, comprising the following steps: Step 1: training data multi-modal acquisition, 12 camera optical motion capture system is used to collect dancer's limb motion data, focusing on capturing real-time angle, motion speed and force moment of 16 key joints; 48kHz sampling rate directional microphone is used to collect training music audio data, and music beat interval, tempo value and accent position are recorded synchronously; Step 2: multi-modal data preprocessing, 5th order Butterworth low-pass filter is used for motion data to filter out noise, missing data is completed by linear interpolation, and abnormal values are removed by 3σ criterion; short-time Fourier transform is used for audio data to extract frequency spectrum characteristics; sliding average filter is used for heart rate data to smooth fluctuations; Step 3: rhythm feature bidirectional extraction, music beat period is identified from audio data by autocorrelation algorithm, dynamic time warping algorithm is used to match music accent moment, and beat intensity weight in each measure is calculated; dancer's key action force moment is extracted from motion data by peak detection algorithm, action period is calculated, and limb coordination degree is analyzed by joint angle change rate; Step 4: synchronization multi-dimension quantification, action and music time difference is calculated; cosine similarity algorithm is used to calculate action sequence and music beat sequence; sliding window is used to calculate synchronization similarity change rate; Step 5: Quantitative results are analyzed by time difference, consistency and dynamic trend data to divide the synchronization into four levels: excellent, good, medium and poor. Weak links are marked on the heat map. The progress of synchronization is calculated by comparing historical training data to generate a progress trend curve. Step 6: Personalized feedback generation. Real-time feedback is superimposed on the rhythm prompt line through AR glasses, with voice prompts. Detailed reports include synchronization level, weak link list, historical comparison chart and improvement suggestions. Training plans are automatically adjusted according to weak links.

[0007] Further, it also includes a motion and music synchronization stability evaluation step, which calculates the synchronization stability coefficient by the formula wherein is the synchronization stability coefficient, t1 is the starting time of the evaluation period, is the time difference between motion and music rhythm at time t. This step calculates for a complete dance segment lasting 1-3 minutes in training, with t1=0s and t2=120s. When <0.6, the time interval of stability decline is automatically located.

[0008] Further, it also includes a multi-dancer collaborative synchronization analysis step, which adds inter-dancer motion synchronization evaluation based on single-person synchronization quantification. Motion capture system is used to collect key joint data of training dancers simultaneously, and the time difference between the same motion of any two dancers is calculated , and the maximum value of in each measure is counted. Cluster algorithm is used to classify multi-dancer motion sequences. A multi-dancer collaborative heat map is generated to mark the weak dancer combinations.

[0009] Further, it also includes a music style adaptability optimization step, which optimizes feature extraction and quantification standards according to the rhythm characteristic differences of different sports dance styles. In waltz style, the music rhythm is 3 / 4 beat, with emphasis on the first beat. The weight of the emphasis beat is set to 1.2, and the weight of the non-emphasis beat is set to 0.6. The motion period needs to match the 3-beat rhythm. In tango style, the music contains split rhythm, and split beat detection is added. In lumbang style, the music tempo is slow, and the matching degree of motion amplitude and music emotion is emphasized.

[0010] Further, it also includes a dynamic synchronization adaptation ability evaluation step, which calculates the synchronization adaptation rate by the formula wherein is the synchronization adaptation rate, S(t) is the synchronization similarity between motion and music at time t, and t is the training time. This step calculates every 10s for a 5-minute evaluation period, and draws a synchronization adaptation rate curve. When When < 0.02, it is determined that the adaptability is weak, and the reason is analyzed.

[0011] Further, it further comprises a training intensity and synchronization correlation analysis step, which combines heart rate data and synchronization quantification results to establish an intensity and synchronization model; the heart rate is divided into three intervals of low intensity, medium intensity and high intensity, and the synchronization scores in each interval are counted; the trend of synchronization with intensity is analyzed, and if the score in the high intensity interval decreases by more than 15%, it is determined that the intensity affects the synchronization; and an intensity and synchronization correlation report is generated, marking the optimal training intensity interval.

[0012] Further, it further comprises a motion detail synchronization optimization step, which, on the basis of key joint motion quantification, increases the synchronization analysis of muscle force and music rhythm; an electromyographic sensor is used to collect the electromyographic signal of the dancer's core muscle, extract the muscle activation time M, and calculate the time difference between M and the music beat time T0 An integral algorithm is used to calculate the muscle activation intensity I, and the matching degree of I and the music beat intensity is analyzed.

[0013] Further, it further comprises a multi-dimensional synchronization comprehensive scoring step, which calculates the comprehensive synchronization score by the formula , wherein is the comprehensive score, is the time difference weight, is the consistency weight, is the stability weight is the time difference standardized score, is the consistency standardized score, is the stability standardized score.

[0014] Further, it further comprises a real-time feedback interaction optimization step, which, on the basis of AR glasses prompts, increases the tactile feedback device; when the synchronization deviation is ≤50ms, the tactile feedback intensity is low; the deviation is 50-100ms, and the tactile feedback intensity is medium; the deviation is >100ms, and the tactile feedback intensity is high, helping the dancer to perceive the synchronization state through tactile perception; at the same time, the voice prompt content is optimized, and the prompt is adjusted in detail according to the dancer's training stage.

[0015] Further, it further comprises a training effect prediction and plan iteration step, which uses an LSTM neural network model to predict the synchronization improvement potential based on the dancer's historical training data; according to the prediction result, the training plan is iterated, if the potential is high, the advanced task is increased; if the potential is medium, the current training intensity is maintained and the weak link training is optimized; if the potential is low, the training method is adjusted; a training effect evaluation report is generated every month.

[0016] Compared with the prior art, the beneficial effects of the present application are: The application solves the problem of subjective ambiguity in traditional training rhythm synchronization judgment through multi-modal data acquisition and quantitative analysis. The system simultaneously collects multi-dimensional data such as actions, audio, heart rate, etc. from time difference, consistency, stability, and other dimensions to quantify synchronization, allowing dancers to clearly understand the details of action and music tempo deviation, the matching degree of action sequence and tempo sequence, and no longer rely on subjective judgment, accurately positioning training weaknesses, and significantly improving training relevance.

[0017] The feedback mechanism of the application has real-time and individualization, and can provide full-process training support for dancers. During training, multi-sensory real-time feedback is achieved through AR glasses, voice, and tactile devices, allowing dancers to immediately perceive synchronization deviations and adjust actions, avoiding "wrong training"; the detailed report generated after training not only includes synchronization level, but also marks specific weaknesses (such as hip joint action deviation in a certain measure, or action period adaptation problem), and gives customized improvement suggestions and targeted training tasks based on historical data, while automatically adjusting subsequent training plans to ensure accurate training direction and avoid ineffective repetition.

[0018] The application can adapt to different training scenarios and needs, expanding the application range of synchronization analysis. For multi-dancer collaborative training, it can quantitatively analyze the synchronization between dancers, mark weak combinations, and generate collaborative training tasks to improve the overall coordination of multi-person dance; for different styles of sports dance, it can automatically identify music style and call corresponding quantitative standards to ensure analysis accuracy; it can also associate training intensity and synchronization to find the optimal training intensity interval for dancers, avoiding synchronization decline due to fatigue, while predicting training effectiveness and iterating plans to match dancer progress pace, reduce ineffective training, enhance dancer training satisfaction, and provide strong support for long-term improvement of sports dance rhythm synchronization. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A schematic block diagram of a quantitative analysis and feedback method for sports dance training rhythm synchronization is proposed for the application; Figure 2 A synchronization multi-dimensional index change line graph for different training stages; Figure 3 A dancer synchronization error comparison bar chart for two-person waltz training; Figure 4 A training intensity and synchronization score correlation scatter plot; Figure 5 A comprehensive synchronization score historical progress line graph. DETAILED DESCRIPTION

[0020] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0021] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0022] In addition, the terms "first", "second" are only for descriptive purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited. In addition, the terms "mounting", "connecting", "connection" should be broadly understood, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium, or internal communication of two elements. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances, and the present application will be further described in detail with reference to the drawings.

[0023] Referring to Figures 1 to 5 A quantitative analysis and feedback method for sports dance training rhythm synchronization, comprising the following steps: Step 1: Multimodal data collection of training data, 12-camera optical motion capture system (frame rate 60 fps, spatial positioning accuracy ±0.5 mm, covering a training area of 10 m x 10 m) is used to collect dancer's limb movement data, focusing on capturing the real-time angles (range 0°-180°), movement speed (0-5 m / s) and force moment of 16 key joints such as hip joint, knee joint and shoulder joint; 48 kHz sampling rate directional microphone is used to collect training music audio data, and music beat interval (50-200 ms), tempo value (60-180 BPM) and accent position (1, 3 beats per bar) are recorded synchronously; wrist heart rate monitor (sampling interval 1 s, accuracy ±1 / min) is used to collect dancer's heart rate data, which is associated with action intensity and rhythm adaptation state, all data are synchronized through time stamp, and synchronization error is ≤10 ms; Step 2: Multimodal data preprocessing, 5th order Butterworth low-pass filter (cutoff frequency 10 Hz) is used for action data to filter out capture noise, and linear interpolation is used to complete the missing data (missing time ≤200 ms) caused by occlusion, and abnormal threshold (±15° beyond normal motion range is judged as abnormal) is set for joint angle, and 3σ criterion is used to remove outliers; Short-time Fourier transform (window length 20 ms, overlap rate 50%) is used for audio data to extract spectral features, and environmental noise is eliminated (signal-to-noise ratio is improved to ≥30 dB); Sliding average filter (window size 5 s) is used for heart rate data to smooth fluctuations and improve data stability, and all preprocessed data are converted to 100 Hz sampling rate time series; Step 3: Rhythm feature extraction in both directions, music beat period T (unit: ms) is identified from audio data by autocorrelation algorithm, dynamic time warping algorithm is used to match music accent moment T0 (unit: s), and intensity weight of each beat in each bar (accent beat weight 1.0, non-accent beat weight 0.7) is calculated; The dancer's key action force moment A (unit: s) is extracted from the action data by peak detection algorithm, and the action period (unit: ms, time interval between adjacent similar forces), joint angle change rate is used to analyze limb coordination C (0-1, the closer to 1, the higher the coordination), and the time corresponding relationship between action features and music features is established; Step 4: Synchronization multi-dimension quantification, calculate action-music time difference Δt (unit: ms, |A-T0|), statistics the average value of Δt in each measure (≤50ms is excellent, 50-100ms is good, 100-150ms is medium, >150ms is poor); calculate the consistency of action sequence and music beat sequence S (0-1, ≥0.8 is high consistency, 0.6-0.8 is medium consistency, <0.6 is low consistency) by using cosine similarity algorithm; analyze the dynamic trend of synchronization with music change, calculate the synchronization similarity change rate by sliding window (window size 3 measures), identify the music paragraph with weak adaptation (change rate < -0.1 is judged as weak paragraph); Step 5: Quantitative result hierarchical analysis, according to time difference, consistency and dynamic trend data, divide the synchronization into four levels: excellent (comprehensive score ≥90 points), good (80-89 points), medium (60-79 points) and poor (<60 points); mark the weak links by heat map, such as "hip joint action in the 2nd-3rd measure is more than 120ms different from the heavy beat" and "turning action period is 20% deviated from the music tempo"; compare the historical training data (the last 10 training), calculate the synchronization progress amplitude (≥10% is significant progress, 5-10% is general progress, <5% is no progress), and generate progress trend curve; Step 6: Personalized feedback generation, real-time feedback superimposes beat prompt line (red for current beat, green for action should reach position) through AR glasses (resolution 1920x1080, field of view 80°), and matches voice prompt ("action a little faster, delay 50ms better"); detailed report includes synchronization level, weak link list, historical comparison chart and improvement suggestion (such as "for the 2nd-3rd measure, it is suggested to practice 20 times slowly first, and focus on the timing of hip joint force"); automatically adjust the training plan according to the weak link, generate targeted task (such as "10 minutes of heavy beat action alignment training every day"), and push to the dancer training APP.

[0024] In the present application, there is also an action-music synchronization stability evaluation step, which is calculated by the formula to calculate the synchronization stability coefficient, wherein is the synchronization stability coefficient (dimensionless, 0-1, ≥0.8 is high stability, 0.6-0.8 is medium stability, <0.6 is low stability), t1 is the starting time of the evaluation period (unit: s), t2 is the end time of the evaluation period (unit: s), is the time difference between action and music beat at t time (unit: ms); this step is for a complete dance segment of 1-3 minutes in training, t1=0s, t2=120s (2 minute segment) are selected for calculation, when When <0.6, the time interval (such as t=40-60s) of automatic positioning stability decline is analyzed, the music features (such as tempo mutation, accent shift) and action features (such as complex turning, fast pace) in the interval are analyzed, and the "weak stability interval special training suggestion" is added in the feedback report, such as "t=40-60s music tempo increases from 120BPM to 130BPM, it is suggested to first segment the action in the interval and gradually improve the tempo adaptation ability".

[0025] In the present application, a multi-dancer coordination synchronization analysis step is also included, which adds inter-dancer action synchronization evaluation on the basis of single-person synchronization quantification; key joint data of 2-5 dancers training coordinately are collected simultaneously by using an action capture system, time difference of the same action between any two dancers is calculated (unit ms, action force time difference of dancer i and dancer j), the maximum value in each measure is counted (≤30ms for excellent coordination, 30-50ms for good coordination, >50ms for poor coordination); a clustering algorithm is used to classify multi-dancer action sequences, and intra-class similarity is calculated (≥0.85 for high coordination, 0.7-0.85 for medium coordination, <0.7 for low coordination); a multi-dancer coordination heat map is generated, and the dancer combination with weak coordination (such as "dancer A and dancer B have poor coordination in turning action") is marked with color; coordination training tasks are added in the feedback, such as "15 minutes of two-person mirror action practice every day, focusing on the force time alignment when turning"; at the same time, coordination training demonstration videos (recorded by professional dancers, with key synchronization points marked) are pushed.

[0026] In the present application, a music style adaptability optimization step is also included, which optimizes feature extraction and quantification standards according to the rhythm feature differences of different sports dance styles (such as waltz, tango, rumba); in waltz style, the music beat is 3 / 4 beat, the accent is in the first beat, the accent beat weight is set to 1.2, the non-accent beat weight is set to 0.6, and the action period needs to match the 3-beat rhythm ); in tango style, the music contains split rhythm, split beat detection is added (identified by spectrum energy mutation), and the time difference threshold of action and split beat is relaxed to ≤70ms; in rumba style, the music tempo is slow (60-90BPM), and the matching degree of action amplitude and music emotion is focused on (through correlation analysis of joint angle change range and music volume change, a correlation coefficient ≥0.7 is high matching); this step automatically identifies the music style (by matching the style database through tempo, time signature, and spectrum features) in the data preprocessing stage, calls the corresponding quantification standard, improves the accuracy of synchronization analysis of different styles of dance, and adds style-specific improvement suggestions in the feedback, such as "in rumba training, it is suggested to increase the action amplitude of the hip joint to improve the matching degree with the music emotion".

[0027] In the present application, a dynamic synchronization adaptation capacity evaluation step is also included, which calculates the synchronization adaptation rate by the formula , wherein is the synchronization adaptation rate (dimensionless, unit s⁻¹, >0 for adaptation improvement, =0 for adaptation stability, <0 for adaptation decline), S(t) is the synchronization similarity between action and music at time t (0-1), t is the training time (unit s); this step takes 5 minutes as an evaluation period, and calculates every 10s , and draws a synchronization adaptation rate curve, when <0.02 for three consecutive 10s, it is determined that the adaptation capacity is weak, and the reasons are analyzed (such as dancer fatigue, music complexity improvement); if it is caused by fatigue, the feedback suggests “shorten the single training time to 30 minutes, and increase the rest interval (rest for 2 minutes every 15 minutes)”; if it is caused by music complexity improvement, low complexity transition music (tempo fluctuation ≤5 BPM, no complex rhythm change) is pushed in the feedback, and it is suggested to improve the adaptation capacity through transition music training first, and then gradually increase the music complexity.

[0028] In the present application, a training intensity and synchronization correlation analysis step is also included, which establishes an intensity-synchronization model by combining heart rate data and synchronization quantization results; the heart rate is divided into low intensity (≤120 times / minute), medium intensity (121-150 times / minute), and high intensity (>150 times / minute) three intervals, and the synchronization scores in each interval are counted (such as low intensity interval score 85 points, medium intensity 82 points, and high intensity 70 points); the trend of synchronization with intensity is analyzed, if the high intensity interval score decreases by more than 15%, it is determined that the intensity affects the synchronization, and the feedback suggests “adjust the training intensity distribution, reduce the high intensity training time proportion from 40% to 25%, and increase the medium and low intensity synchronization consolidation training”; at the same time, an intensity-synchronization correlation report is generated, which marks the optimal training intensity interval (such as “the optimal intensity of the dancer is 121-150 times / minute, and the synchronization score is the highest”), helps the dancer to reasonably plan the training intensity, and reduces the synchronization decline caused by excessive fatigue.

[0029] In the present application, a motion detail synchronization optimization step is also included, which increases the synchronization analysis of muscle force and music rhythm on the basis of key joint motion quantization; an electromyographic sensor (sampling rate 1000 Hz, resolution 12 bits) is used to collect the electromyographic signals of 8 core muscles of the dancer's thigh quadriceps and calf gastrocnemius, extract the muscle activation time M (unit s), and calculate the time difference between M and the music beat time T0 (unit ms, ≤30 ms is excellent, 30-50 ms is good, >50 ms is poor); integral algorithm is used to calculate muscle activation intensity I (unit mV·s, positively correlated with movement strength), and the matching degree of I and music beat intensity (correlation coefficient ≥0.6 is high matching) is analyzed; muscle power improvement suggestions are added in the feedback, such as "the activation time of the quadriceps muscle lags behind the beat by 25 ms, it is suggested to advance the power time, which can be felt through slow motion mirror practice", and electromyogram-beat synchronization training animation (annotating the corresponding relationship between muscle activation and beat) is pushed.

[0030] In the present application, a multi-dimensional synchronization comprehensive score step is also included, which is calculated by the formula to calculate the comprehensive synchronization score, wherein is the comprehensive score (0-100 points), is the time difference weight (value 0.4), is the consistency weight (value 0.3), is the stability weight (value 0.3) is the time difference standard score (0-100 points, Δt≤50 ms is 100 points, and decreases by 25 points for each increase of 50 ms), is the consistency standard score (0-100 points, S≥0.8 is 100 points, and decreases by 15 points for each decrease of 0.1), is the stability standard score (0-100 points, ≥0.8 is 100 points, and decreases by 15 points for each decrease of 0.1); this step automatically collects the data of each dimension to calculate the score, divides the level according to the score and matches the corresponding training scheme (90-100 points push advanced skill training, 80-89 points push consolidation training, 60-79 points push basic synchronization training, and <60 points push entry rhythm perception training), and generates a comprehensive score history curve to intuitively show the dancer's synchronization improvement trajectory and enhance the training sense of achievement.

[0031] In the present application, a real-time feedback interaction optimization step is also included, which increases a tactile feedback device (worn on the wrist, vibration frequency consistent with the music beat, and intensity positively correlated with the synchronization deviation) based on the AR glasses prompt; when the synchronization deviation is ≤ 50 ms, the tactile feedback intensity is low (vibration amplitude 0.1 mm); when the deviation is 50-100 ms, it is medium (0.2 mm); and when the deviation is > 100 ms, it is high (0.3 mm), helping the dancer to perceive the synchronization state through tactile sensation, especially in noisy music or obstructed vision scenarios to improve feedback effectiveness; at the same time, the voice prompt content is optimized, and the prompt is adjusted in detail according to the dancer's training stage (in the primary stage, the prompt is "left foot forward, align with the first beat", and in the advanced stage, the prompt is "hip joint force lag, adjust 0.03 s"); the feedback interaction confirmation function is increased, and the dancer can pause / continue feedback or switch feedback mode (AR+voice / tactile alone mode) through gestures (such as waving hands), improving training autonomy.

[0032] In the present application, a training effect prediction and plan iteration step is also included, which uses an LSTM neural network model to predict the synchronization improvement potential in the next month (such as "expected to improve the comprehensive score from 75 points to 85 points in 2 weeks") based on the dancer's historical training data (synchronization score, weak link, training duration of the last 30 training sessions); according to the prediction result, the training plan is iterated, if the potential is high (improve ≥ 10 points), add advanced tasks (such as "learn to synchronize complex rhythm type actions"); if the potential is medium (5-10 points), maintain the current training intensity and optimize the weak link training; if the potential is low (< 5 points), adjust the training method (such as "change the music style and use a more familiar rhythm type to improve interest"); generate a training effect evaluation report every month to compare the predicted value with the actual value (deviation ≤ 5 points is accurate prediction), optimize the neural network model parameters, and improve the prediction accuracy of subsequent predictions to ensure that the training plan always fits the dancer's progress pace and reduces ineffective training.

[0033] The specific embodiments of the present application are further illustrated by the following two examples: Example 1: Primary dancer single lumba training rhythm synchronization analysis scene This example is aimed at a primary lumba dancer (training duration of 6 months, mastered basic steps but rhythm synchronization is unstable), who performs 45 minutes of single lumba training in a 10m x 10m dance training room, to solve the problems of large deviation between hip joint action and music accent, decreased synchronization after high-intensity training, and inability to accurately locate weak links. The specific implementation process is as follows: I. System deployment and parameter configuration Multi-modal acquisition device layout: The motion capture system uses 12 OptiTrack Prime 17W cameras, which are distributed in a ring around the training room (height 2.5 m, interval 2 m). The frame rate is set to 60 fps, and the spatial positioning accuracy is ±0.5 mm. The key joints of the dancer are marked, including the hip joint, knee joint, ankle joint, shoulder joint, and elbow joint. The joint angle acquisition range is 0°-180°, and the motion speed acquisition range is 0-3 m / s (the speed of the rumba dance is relatively slow). Two 48 kHz directional microphones are fixed on the front and back walls of the training room (3 m away from the dancer), respectively, to collect rumba training music (tempo 75 BPM, 4 / 4 beat, emphasis on the first beat), with an audio sampling interval of 2.1 ms. The dancer wears a heart rate monitor (sampling interval 1 s, accuracy ±1 / min), which is synchronized with the training APP. All devices are synchronized by timestamp, and the synchronization error is calibrated to 8 ms.

[0034] Core parameter preset: In the data preprocessing stage, the motion data is filtered by a 5th order Butterworth low-pass filter (cutoff frequency 10 Hz), and the threshold for linear interpolation to complete the missing data is set to 200 ms (rumba action is continuous, and the occlusion time is short). The abnormal threshold of joint angle is set to ±15° of the normal motion range (for example, the normal motion range of the hip joint is 0°-120°, and the abnormal value is determined as <-15° or >135°). The window length of the short-time Fourier transform of the audio data is 20 ms, the overlap rate is 50%, and the signal-to-noise ratio is improved by 35 dB. The sliding average filter window of the heart rate data is 5 s, and the heart rate interval after smoothing is divided into: low intensity ≤120 beats / min, medium intensity 121-150 beats / min, and high intensity >150 beats / min. The AR glasses use Microsoft HoloLens 2 (resolution 1920x1080, field of view 80°), the voice prompt volume is 60 dB, and the vibration amplitude of the tactile feedback device (worn on the left wrist) is classified as: low 0.1 mm, medium 0.2 mm, and high 0.3 mm.

[0035] II. Core training process and formula application Data acquisition and preprocessing: After the start of training, the motion capture system collects the joint data of the dancer's rumba basic steps (cucaracha, fan step, alemana) in real time. At the 5th minute, the dancer performs the fan step, and the hip joint is occluded by the skirt for 150 ms, which is completed by linear interpolation (the angle deviation after completion is <2°). The microphone collects music audio, and after short-time Fourier transform, the air conditioner noise in the training room is filtered out (the signal-to-noise ratio is improved from 22 dB to 36 dB). The heart rate monitor collects data, and after sliding average, the initial heart rate is displayed as 85 beats / min (low intensity). After preprocessing, all data are converted into time series with a sampling rate of 100 Hz and stored in the local server.

[0036] Rhythm feature extraction and synchronicity quantification: From audio data, the self-correlation algorithm identifies the rhythm period of the rumba T = 800 ms (75 BPM = 60000 ms / 75 = 800 ms), and the dynamic time warping algorithm matches the accent time T0 (e.g. 10s, 10.8s, 11.6s), with accent beat weight 1.0 and non-accent beat weight 0.7. From the action data, the peak detection algorithm extracts the hip joint force time A (e.g. 10.05s, 10.83s, 11.62s), and calculates the action period = 780 ms (close to T = 800 ms), and the joint angle change rate analyzes the limb coordination C = 0.75 (medium coordination). Synchronicity quantification shows: the average value of action-music time difference Δt is 65 ms (good), the consistency of action sequence and music rhythm sequence S = 0.72 (medium consistency), and the sliding window (3 bars) calculates the synchronization similarity change rate -0.05 (not up to the weak section standard).

[0037] Synchronization stability and dynamic adaptability evaluation: After 20 minutes of training, select the complete rumba segment at t1 = 600 s (10 minutes) and t2 = 720 s (12 minutes) to calculate the synchronization stability coefficient. Given that the average value of Δt(t) in this period is 60 ms, the maximum is 80 ms, and the minimum is 40 ms, substitute into the formula The integral result is the time integral of Δt(t). Assuming the average value of Δt(t) is 60 ms, the integral , and substituting gives (high stability). When training for 30 minutes, the dancer's heart rate rises to 155 beats per minute (high intensity), and the synchronization adaptation rate is calculated every 10 s, and the continuous three times are -0.025, -0.03, and -0.028 (all < -0.02), which is judged as weak adaptability, and the reason is fatigue, and the feedback suggestion is "shorten the single training to 30 minutes, rest for 2 minutes and continue, and the proportion of subsequent high-intensity training is reduced to 25%".

[0038] Personalized feedback and plan adjustment: In real-time feedback, the AR glasses superimpose the current beat line (1st beat) in red and the action should reach the position line in green, showing "hip joint force lag 20 ms, align with the 1st beat"; the tactile feedback device vibrates at an amplitude of 0.2 mm (medium intensity); the voice prompt is "action is slightly slow, delay 20 ms better". After training, a report is generated: synchronization level good (82 points), weak links are "15-18 small section fan-shaped step hip joint and strong beat tempo difference 85 ms" "high intensity interval (heart rate > 150 times / minute) synchronization score decreased by 18%"; compared with the last 10 training, the progress rate is 8% (general progress); automatically adjust the training plan, generate "10 minutes of strong beat tempo hip joint alignment training every day" "high intensity training time controlled within 10 minutes" tasks, and push to the dancer APP.

[0039] III. Training effect data representation Table 1: Comparison of key indicators of single person rumba training in example 1 In table 1: the data reflects the adaptability of the present application to the initial single person rumba training: the synchronization comprehensive score is improved from 65 points to 82 points, which is due to the accurate positioning of the problem by multi-dimensional quantitative analysis; the weak link positioning only takes 5 minutes, which is greatly shortened compared with the traditional 30 minutes, so that the dancer can quickly focus on the hip joint deviation problem; the real-time feedback delay is 25 ms, which avoids the "wrong training" caused by finding the problem after traditional training; the decrease amplitude of high intensity synchronization is reduced from 35% to 18%, because the system timely identifies fatigue and adjusts the training intensity; the progress rate is 12% after 1 week, which is much higher than the traditional 5%, proving the effectiveness of the personalized training plan, and solving the pain points of unstable rhythm synchronization and difficult problem positioning of the initial dancer.

[0040] Example 2: Rhythm synchronization analysis scene of intermediate dancer double person waltz cooperative training This embodiment is aimed at two intermediate waltz dancers (partner training for 8 months, mastering rotation, swing and other actions, with deviation problem in cooperative turning), and 60 minutes of double person waltz training is carried out in a 12m x 12m training room, aiming to solve the problems of large cooperative synchronization error between dancers, inaccurate waltz 3 / 4 beat rhythm adaptation and fuzzy comprehensive synchronization score, and the specific implementation process is as follows: I. System deployment and parameter configuration Multi-device collaborative acquisition layout: The motion capture system uses 16 OptiTrack Prime41 cameras (frame rate 60fps, accuracy ±0.5mm), covering a two-person training area (12m x 12m), marking the hip joints, shoulder joints, ankle joints, etc. of two dancers (Dancer A, Dancer B), focusing on monitoring the shoulder joint angle (0°-360°) in the rotating action; 3 48kHz directional microphones are placed in a triangular shape (2-4m away from the two-person combination), collecting waltz music (tempo 120BPM, 3 / 4 beat, emphasis on the first beat); Both dancers wear heart rate monitors (sampling interval 1s); The downhole-ground communication module uses a 5G private network (transmission rate 100Mbps, delay 20ms) to ensure synchronous transmission of two-person data; AR glasses are worn by the two dancers, displaying a two-person action synchronization prompt line (blue for Dancer A's action, pink for Dancer B's action).

[0041] Style adaptation and scoring parameter preset: In the music style adaptation stage, the system automatically identifies the waltz 3 / 4 beat characteristics, sets the emphasis beat weight to 1.2, the non-emphasis beat weight to 0.6, and the action period ; Comprehensive scoring parameters: w1=0.4 (time difference weight), w2=0.3 (consistency weight), w3=0.3 (stability weight), (time difference standard score) ≥90 points corresponds to Δt≤30ms, (consistency standard score) ≥90 points corresponds to S≥0.85, (stability standard score) ≥90 points corresponds to ≥0.85; In multi-dancer collaborative evaluation, (time difference between dancers) excellent threshold ≤30ms, intra-class similarity high collaboration threshold ≥0.85.

[0042] II. Core collaborative process and formula application Multi-dancer data acquisition and collaborative analysis: After the training starts, the motion capture system synchronously collects the waltz basic action (right turn, left turn, cross step) data of Dancer A and Dancer B, and at the 15th minute, the right turn action is calculated. Dancer A's shoulder joint force moment A1=450.2s, Dancer B's shoulder joint force moment A2=450.05s =150ms-50ms=100ms? Corrected to =|450.2-450.05|=150ms (>50ms, poor collaboration); Use K-means clustering algorithm to classify the two-person action sequence, intra-class similarity 0.72 (medium collaboration); Generate a collaborative heat map, mark "Dancer A and Dancer B right turn action collaboration difference", and AR glasses display "Dancer B's action is 150ms ahead, adjust the force moment to 450.2s".

[0043] Overall score and plan iteration: At the 30-minute training, the overall synchronization score is calculated. It is known = 85 points (Δt average 40 ms), = 80 points (S = 0.82), = 88 points ( = 0.88), into the formula points (good); compared with the last 10 collaborative training, the overall score improved from 75 points to 84.4 points, an improvement of 12.5% (significant progress). The training effect prediction module is based on the LSTM model, which inputs the last 30 training data (scores 75-84.4, weak links are right turn coordination, training duration 60 minutes / time), predicts that the overall score in the next two weeks can reach 90 points (excellent); after plan iteration, increase "15 minutes of daily two-person right turn mirror practice, focus on shoulder joint force moment" task, adjust the waltz music tempo from 120 BPM to 125 BPM, and enhance the rhythm adaptation difficulty.

[0044] Dynamic adaptation and feedback optimization: At the 45-minute training, dancer A's heart rate is 145 beats per minute (moderate intensity), and dancer B's heart rate is 152 beats per minute (high intensity), the synchronization adaptation rate is calculated , dancer A's continuous 3 times = 0.01, 0.005, 0.008 (> 0, adaptation improvement), dancer B's continuous 3 times = -0.022, -0.025, -0.021 (< -0.02, adaptation decline), the reason is that dancer B has been training at high intensity for too long, and the feedback suggests "dancer B rest for 3 minutes, and reduce the proportion of high-intensity training to 20%"; the tactile feedback device vibrates 0.3mm (high) for dancer B, prompting "synchronization of the action is decreasing, adjust the intensity".

[0045] III. Collaborative training effect data representation Table 2: Comparison of key indicators of two-person waltz collaborative training in Example 2 In Table 2: The data reflects the adaptability of the invention to two-person waltz collaborative training: the synchronization error between dancers is reduced from 120 ms to 45 ms, the intra-class similarity is improved from 0.65 to 0.83, solving the low-efficiency problem of "relying on tacit understanding"; the overall score improvement reaches 12.5%, far exceeding the traditional 4%, proving the effectiveness of quantitative analysis and personalized planning; the collaborative weak positioning only takes 3 minutes, which is much shorter than the traditional 45 minutes, allowing the two people to quickly focus on the right turn action coordination problem; after plan iteration, the predicted score reaches 90 points, providing a clear goal for future training and avoiding blind practice. Overall, the system effectively improves the precision and efficiency of two-person collaborative training, and promotes intermediate dancers from "knowing how to dance" to "dancing well".

[0046] Referring to Figure 2 : The figure intuitively presents the long-term progress trend of the dancer's multi-dimensional synchronization indicators. With the increase of training times, the action-music time difference continues to shorten from 95ms to 35ms, indicating that the time alignment accuracy of action and rhythm has been significantly improved; the consistency and stability coefficients increase from about 0.6 to more than 0.85, reflecting that the matching degree of action sequence and music rhythm and the persistence of synchronization state are greatly optimized. The chart can help dancers and coaches clearly see the improvement rhythm of each dimension and judge the effectiveness of the training plan, such as the fastest decline in time difference in the 10th-15th training, which can analyze the advantages of the training method (such as special alignment exercises) in this stage, and provide a basis for subsequent training adjustment.

[0047] Referring to Figure 3 : The figure clearly shows the advantages of the invention in double-person collaborative training, and the synchronization error of all action types is greatly reduced, especially for complex actions such as right turn and left turn rotation, the error is reduced from more than 100ms to about 30ms, reaching the excellent collaboration standard (≤30ms). Traditional tacit grinding relies on the subjective feeling of dancers, with large and unstable errors, while the invention accurately solves the weak points of collaboration by quantifying the action time difference between dancers and generating targeted collaborative tasks (such as right turn rotation force alignment exercises). For example, the error of the swing step is reduced from 120ms to 35ms, making the double-person action more unified and improving the smoothness of dance performance, while shortening the collaborative training period and avoiding long-term ineffective grinding.

[0048] Referring to Figure 4 : The figure intuitively shows the influence of training intensity on synchronization through scatter points and trend lines. When the heart rate is 120-140 times / minute (medium intensity), the synchronization score is the highest (85-88 points), at this time the dancer is full of energy, which can ensure the quality of action and accurately perceive the rhythm; when the heart rate is <120 times / minute (low intensity), the score is lower (78-85 points), because the action intensity is insufficient and the rhythm perception is not sensitive; when the heart rate is >150 times / minute (high intensity), the score decreases significantly (62-75 points), because fatigue causes action deformation and force timing deviation. The chart helps dancers find the optimal training intensity interval and avoid blindly increasing intensity, such as reducing the proportion of high-intensity training from 40% to 25%, increasing the synchronization consolidation training of medium intensity, and improving the overall training efficiency.

[0049] Referring to Figure 5The figure embodies the training effect prediction and history progress tracking function of the present application. The actual score steadily increases from 65 points in the first week to 85 points in the sixth week, and the overall trend is upward, with a slight decrease in the fourth week (73 points). The reason can be analyzed in combination with the training record (such as the complexity of music increasing in that week); the deviation between the predicted score and the actual score is small (≤2 points), indicating that the prediction model is accurate and can provide clear progress expectations for dancers. For example, the prediction score is 88 points in the sixth week, which encourages dancers to continue training according to the plan; at the same time, the chart helps the coach verify the effectiveness of the training plan, such as the rapid increase in scores from the third to the fifth week due to the addition of heavy accent beat special training. This training method can be continued in the future to enhance the training confidence and sense of purpose of the dancers.

[0050] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can make equivalent replacements or changes within the technical range disclosed by the present application according to the technical solution and inventive concept of the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for quantitative analysis and feedback of the synchronization of the rhythm in sports dance training, characterized by, Comprising the following steps: Step 1: Multi-modal acquisition of training data, using a 12-camera optical motion capture system to collect dancer's limb movement data, focusing on capturing real-time angles, movement speed and force timing of 16 key joints; using a 48kHz sampling rate directional microphone to collect training music audio data, synchronously recording music beat interval, tempo value and accent position; Step 2: Multi-modal data preprocessing, using a 5th order Butterworth low-pass filter to filter out noise from the motion data, filling in missing data through linear interpolation, and removing outliers using the 3σ criterion; using short-time Fourier transform to extract spectral features from audio data; using sliding average filter to smooth fluctuations in heart rate data; Step 3: Rhythm feature extraction in both directions, identifying music beat period from audio data using autocorrelation algorithm, matching music accent timing using dynamic time warping algorithm, calculating the intensity weight of each beat in each measure; extracting dancer's key action force timing from motion data using peak detection algorithm, calculating action period, and analyzing limb coordination using joint angle change rate; Step 4: Synchronicity multi-dimensional quantification, calculating the time difference between action and music; using cosine similarity algorithm to calculate the consistency of action sequence and music beat sequence; calculating the synchronization similarity change rate through sliding window; Step 5: Hierarchical analysis of quantification results, dividing synchronicity into four levels of excellent, good, medium and poor according to time difference, consistency and dynamic trend data; marking weak links through heat map; comparing with historical training data to calculate the progress of synchronicity, generating progress trend curve; Step 6: Generation of personalized feedback, real-time feedback through AR glasses superimposed with beat prompt line, accompanied by voice prompt; Detailed report includes synchronicity level, weak link list, historical comparison chart and improvement suggestions; automatically adjust training plan according to weak links.

2. The method according to claim 1, wherein, Also included is a movement, music synchronization stability evaluation step, which is performed by the formula calculating a synchronization stability coefficient, wherein is the synchronization stability coefficient, t1 is the start time of the evaluation period, is the time difference between the movement and the music beat at time t; this step is performed for a complete dance segment lasting 1-3 minutes in training, with t1 = 0 s, t2 = 120 s, and when <0.6, automatically positioning the time interval in which stability is reduced.

3. The method of claim 1, wherein the method further comprises: It also includes a multi-dancer collaboration synchronization analysis step, which, based on the quantification of individual synchronization, adds an assessment of the synchronization of movements between dancers; using a motion capture system to simultaneously collect key joint data of the trained dancers, and calculating the time difference of similar movements between any two dancers. Statistics within each section The maximum value; a clustering algorithm is used to classify the movement sequences of multiple dancers; a multi-dancer collaboration heatmap is generated, and dancer combinations with weak collaboration are marked with colors.

4. The method of claim 1, wherein the method further comprises: It also includes a music style adaptability optimization step, which optimizes feature extraction and quantification standards according to the rhythm feature differences of different sports dance styles; in waltz style, the music beat is 3 / 4, the accent is in the first beat, the accent beat weight is set to 1.2, the non-accent beat weight is 0.6, and the action period needs to match the 3-beat rhythm; in tango style, the music contains split rhythm, split beat detection is added; in lumbang style, the music tempo is slow, focusing on quantifying the matching degree of action amplitude and music emotion.

5. The method for quantitative analysis and feedback of rhythm synchronicity in sports dance training according to claim 1, characterized in that, Also included is a dynamic synchronization adaptation capacity evaluation step, which calculates the synchronization adaptation rate by the formula S(t) = (S(t) - S(t-1)) / S(t-1) S(t) is the synchronization adaptation rate, S(t) is the synchronization similarity between action and music at time t, t is the training time; this step is an evaluation period of 5 minutes, and is calculated every 10s , draw the synchronization adaptation rate curve, when the synchronization adaptation rate is less than -0.02 for 3 consecutive 10s , it is determined that the adaptation capacity is weak, and the reason is analyzed.

6. The method for quantitative analysis and feedback of rhythm synchronicity in sports dance training according to claim 1, characterized in that, It also includes a training intensity and synchronicity correlation analysis step, which combines heart rate data and synchronicity quantification results to establish an intensity and synchronicity model; dividing heart rate into low, medium and high intensity intervals, and calculating the synchronicity score in each interval; Analyze the trend of synchronicity with intensity, if the score in high intensity interval decreases by more than 15%, determine that intensity affects synchronicity; at the same time, generate an intensity and synchronicity correlation report.

7. The method of claim 1, wherein the method further comprises: Further comprising action detail synchronization optimization step, which is based on key joint action quantification, increases the synchronization analysis of muscle force and music rhythm; adopts electromyographic sensor to collect the electromyographic signal of dancer's core muscle, extracts muscle activation time M, and calculates the time difference between M and music beat time T0 .

8. The method for quantitative analysis and feedback of rhythm synchronicity in sports dance training according to claim 1, characterized in that, Also included is a multi-dimension synchronicity composite scoring step that calculates a composite synchronicity score by the formula wherein is the composite score, is the time difference weight, is the consistency weight, is the stability weight is the time difference normalized score, is the consistency normalized score, is the stability normalized score.

9. The method for quantitative analysis and feedback of rhythm synchronicity in sports dance training according to claim 1, characterized in that, It also includes a real-time feedback interaction optimization step, which adds a haptic feedback device based on AR glasses prompt; when the synchronization deviation is ≤50ms, the haptic feedback intensity is low; when the deviation is 50-100ms, it is medium; when the deviation is >100ms, it is high, helping the dancer to perceive the synchronization state through touch; at the same time, optimize the voice prompt content.

10. The method for quantitative analysis and feedback of rhythm synchronicity in sports dance training according to claim 1, characterized in that, It also includes training effect prediction and planning iteration steps, which use an LSTM neural network model based on the dancer's historical training data to predict the potential for synchronization improvement. Based on the prediction results, the training plan is iterated. If the potential is high, advanced tasks are added; If the potential is medium, maintain the current training intensity and optimize weak link training; if the potential is low, adjust the training method; A monthly training effect evaluation report is generated.

Images