Intelligent badminton action analysis system and method based on motion sensor

Abstract

The invention belongs to the technical field of exercise training and analysis, and particularly relates to an intelligent badminton action analysis system and method based on a motion sensor, and the system comprises a sensor data collection module, a real-time data processing module, an action recognition analysis engine, an intelligent data storage module and a visualization module. The method comprises the following steps: receiving and buffering sensor data in real time; periodically extracting data segments by using a sliding window to carry out multi-feature (an acceleration peak value, an angular velocity peak value, a peak value ratio and the like) calculation; based on a predefined multi-feature threshold rule set, identifying action types (such as kicking, flat ball drawing and the like) and evaluating the quality; an intelligent strategy is adopted to selectively store data so as to optimize the system load; and finally, a result is visually fed back in real time. The method realizes objective, quantitative and real-time analysis of badminton actions, has the advantages of high recognition accuracy, high system real-time performance, efficient data processing and the like, and is suitable for autonomous training of athletes and auxiliary guidance of coaches.

Description

Technical Field

[0001] This invention relates to the field of sports training and analysis technology, and in particular to an intelligent badminton motion analysis system and method based on motion sensors. Background Technology

[0002] Badminton is a sport that demands a high degree of technical precision. Improving an athlete's skill level relies on long-term, systematic, and timely feedback-based specialized training. Traditional badminton training methods primarily depend on on-site observation and experience-based guidance from coaches, which has the following significant limitations: Poor real-time feedback: Trainees cannot immediately know the quality of their movements during practice. Coaches usually need to provide feedback during training breaks or after the training session, which cannot achieve "correction during practice" and results in low training efficiency.

[0003] The evaluation is highly subjective: coaches' judgments on the merits of movements are mostly based on personal experience, lacking objective and quantitative evaluation standards. It is difficult to accurately evaluate the details of the movements (such as the explosiveness of power and the stability of the swing trajectory), and it is easily affected by the coach's state and perspective.

[0004] Data is difficult to retain and analyze: The training process lacks systematic data recording, making it difficult to track, compare, and analyze the historical training data of trainees, and thus unable to provide data support for the development of long-term, personalized training plans.

[0005] Limited personalized guidance: When one coach guides multiple students at the same time, it is difficult to conduct in-depth analysis and customize personalized solutions for each student's unique technical problems.

[0006] Cost and resource constraints: The scarcity of high-level coaches and the high cost of hiring them long-term limit the accessibility and sustainability of training.

[0007] With the development of sensor technology, some solutions for motion monitoring using inertial measurement units (IMUs) have emerged. However, when applied to high-speed sports like badminton with diverse movement types, existing technologies typically face the following challenges: high data processing latency, failing to meet real-time feedback requirements; simple motion recognition models, insufficient ability to distinguish between similar but subtly different movements such as smashes, drop shots, flat drives, and high clears; lack of efficient storage and management strategies for high-frequency sensor data streams, making the system prone to crashing due to data surges; and low overall integration, failing to form a complete closed loop from data acquisition to intelligent analysis and then to visual feedback.

[0008] Therefore, there is an urgent need to develop a system and method that can accurately and with low latency identify and analyze badminton movements in real time, effectively manage massive amounts of sensor data, and ultimately provide intuitive and intelligent training guidance. Summary of the Invention

[0009] Based on the technical problems existing in the prior art, this invention proposes an intelligent badminton motion analysis system and method based on motion sensors.

[0010] The intelligent badminton motion analysis system based on motion sensors proposed in this invention includes: Sensor data acquisition module: configured to receive data streams sent by motion sensors fixed on badminton rackets in real time via wireless communication protocol (preferably UDP protocol); the sensor is preferably an IMU device integrating a three-axis accelerometer and a three-axis gyroscope (such as WT901WIFI type), and the data acquisition frequency is not less than 100Hz.

[0011] Real-time data processing and buffering module: It adopts a sliding window mechanism to manage the data stream of each sensor; this module maintains a circular buffer to temporarily store the raw sensor data of the most recent period (e.g., corresponding to 400 data points, about 4 seconds); this module is also responsible for the preliminary parsing and verification of the data.

[0012] Action recognition and analysis engine: This is the core of the invention, comprising: Feature extraction unit: Calculates multidimensional feature values ​​from the data in the sliding window, including but not limited to: peak acceleration, average acceleration, peak angular velocity, average angular velocity, and peak ratio (defined as the ratio of peak acceleration to average acceleration), which characterizes explosive force.

[0013] Action classification unit: Based on predefined multi-feature threshold rules, the extracted features are mapped to specific badminton action types, for example: Smash: The criteria for recognition are peakAcceleration>12g, maxGyro>800° / s, and peakRatio>2.2.

[0014] Flat drive (Drive): The identification conditions are peakAcceleration>7g, maxGyro>500° / s, and peakRatio>1.5.

[0015] High Clear: The recognition criteria are peakAcceleration > 6g and avgAcceleration > 3g and 1.2g. <peakRatio<2.0。

[0016] Lift: The recognition conditions are peakAcceleration>3g and maxGyro<600° / s.

[0017] Quality assessment unit: Based on the identified action type and the calculated relevant parameters (such as the estimated swing speed), the quality of action execution is graded (e.g., "excellent", "good", "average").

[0018] Data storage and management module: Implements intelligent data sampling strategies to cope with massive amounts of data. The strategies include: storing 100% of data marked as "keyframes" (such as action start / end points) and data containing valid action recognition results; downsampling and storing ordinary continuous sampling data at a fixed ratio (such as 10%); this module is also responsible for automatically cleaning up expired historical data periodically.

[0019] Intelligent Analysis and Suggestion Module (Optional): Integrates or calls AI models (such as the DeepSeek model) to deeply mine the accumulated action analysis results and generate personalized training reports, technical shortcomings analysis and targeted improvement suggestions.

[0020] Data visualization and interaction module: Provides a graphical user interface for real-time display of sensor status, motion recognition results, speed curves, quality assessment dashboards, historical data statistical charts, and training leaderboards.

[0021] System Support Modules: These include backend support functions such as user permission management, multi-sensor device management, and course and training plan management.

[0022] Accordingly, the present invention also provides an intelligent badminton motion analysis method based on motion sensors, comprising the following steps: S1: Receives raw data packets in real time from the motion sensor attached to the badminton racket via a wireless network.

[0023] S2: Parse the data packet and add the parsed data to the sliding window (circular buffer) corresponding to the sensor.

[0024] S3: Periodically or when triggering conditions are met (e.g., when the amount of data in the sliding window reaches a certain threshold and is a multiple of a specific analysis period), perform action recognition analysis on the latest continuous data segment (e.g., the most recent 400 data points) in the window.

[0025] S4: The action recognition and analysis sub-step includes: S4.1: Calculate the multi-dimensional motion characteristics of this data segment, including the peak and mean values ​​of the composite acceleration, the peak and mean values ​​of the composite angular velocity, the ratio of peak acceleration to mean value, and estimate the swing speed; S4.2: Match the calculated feature values ​​with the predefined threshold rules for each action type to determine the action type corresponding to the current data segment; S4.3: Based on the determined action type and estimated speed value, evaluate the execution quality of the action by comparing it with the preset quality standard.

[0026] S5: Based on the intelligent sampling strategy, decide whether to save the current raw data and / or motion analysis results to the persistent database. The strategy prioritizes saving keyframe data and motion recognition results.

[0027] S6: Push the motion recognition and quality assessment results to the front-end visualization interface for display in real time.

[0028] S7: (Optional) Periodically or as needed, invoke the AI ​​analysis model to generate comprehensive training analysis and suggestions based on the user's historical action data.

[0029] Compared with existing technologies, this invention provides an intelligent badminton motion analysis system and method based on motion sensors, which has the following beneficial effects: High real-time performance: Through sliding window and asynchronous processing technology, real-time processing and analysis of 100Hz high-frequency sensor data is realized. The end-to-end delay from action recognition to result display can be controlled within 1 second, meeting the needs of real-time feedback in training.

[0030] High recognition accuracy: The rule recognition method adopts multi-feature fusion, which comprehensively considers the magnitude of acceleration, angular velocity change and explosive force characteristics (peak ratio). It can effectively distinguish badminton techniques with similar characteristics but different mechanical modes, such as smash, flat drive, high clear, and lob. Experiments show that the average recognition accuracy can reach more than 85%.

[0031] Excellent system performance: The innovative intelligent data sampling strategy reduces the amount of data that needs to be permanently stored by about 90% while ensuring that all critical action information is not lost. This greatly alleviates the storage and I / O pressure on the database and improves the overall stability and scalability of the system.

[0032] The architecture is highly scalable: it adopts a modular design, supports multiple sensor terminals to be connected and processed independently at the same time, and the modules are decoupled through clear interfaces, which facilitates subsequent algorithm upgrades or function expansion.

[0033] Intelligent assistance: By integrating an AI analysis module, it can go beyond single action evaluation, and mine users' technical patterns and progress trends from long-term data to provide more in-depth personalized training guidance.

[0034] Excellent user experience: Rich visualization components transform abstract sensor data into intuitive charts, scores, and rankings, significantly enhancing the scientific rigor and enjoyment of training.

[0035] Lowering costs and barriers to entry: This system can assist or even partially replace the coach's real-time observation and evaluation functions, enabling athletes to conduct more efficient self-training and reducing the economic costs and resource barriers of professional training. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the overall architecture of an embodiment of the system of the present invention; Figure 2 This is a schematic diagram of the sliding window processing in the action recognition process of the present invention; Figure 3 This is a flowchart of the intelligent data sampling strategy of the present invention; Figure 4 This is a schematic diagram of the feature value calculation process in the action recognition of the present invention. Detailed Implementation

[0037] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0038] Example 1: System Architecture and Deployment like Figure 1 As shown, the intelligent badminton motion analysis system of the present invention adopts a layered architecture design, including: Sensor device layer: Composed of multiple WT901WIFI motion sensors, which are securely fixed to specific positions on the badminton racket handle or frame with straps. The sensors collect data such as acceleration and angular velocity at a frequency of 100Hz and send 54-byte binary data packets to a specified port (such as 1399) of the server via Wi-Fi using the UDP protocol.

[0039] Data access and processing layer: Deployed on the server, it includes a UDP packet receiving service, a data parser, and a sliding window manager based on a circular buffer. This layer is responsible for converting the raw byte stream into structured sensor data objects and storing them in the corresponding sliding window.

[0040] Business logic layer: Includes core action recognition engine, intelligent sampling decision-maker, and asynchronous task scheduler. The action recognition engine obtains data from the sliding window and performs feature calculation and classification. The sampling decision-maker determines the storage destination of the data. All time-consuming database operations are executed through asynchronous tasks to avoid blocking the real-time data stream.

[0041] Data storage layer: Use a relational database (such as MySQL) to store sampled sensor metadata, action recognition results, user information, etc., and use an in-memory database (such as Redis) to cache hot data and session information.

[0042] Application service layer: Provides RESTful API or WebSocket interface for front-end to call, for querying real-time analysis results, historical data, receiving AI suggestions, etc.

[0043] User Presentation Layer: This is the web front-end application, including student, coach, and administrator clients, displaying all information in the form of charts, dashboards, lists, etc.

[0044] Example 2: Sliding Window and Action Recognition Process like Figure 2 As shown, the specific process of real-time action recognition is as follows: The UDP service continuously receives data packets, which are then parsed to obtain SensorData objects containing fields such as timestamp, triaxial acceleration (ax, ay, az), and triaxial angular velocity (gx, gy, gz).

[0045] Add the object to the CircularBuffer instance assigned to this sensor ID. This buffer has a capacity of 400, and when it is full, new data will overwrite the oldest data.

[0046] Set the recognition trigger condition: whenever the amount of data in the buffer reaches or exceeds 200, and the sequence number of the currently processed data packet is a multiple of 50, start an action recognition analysis.

[0047] During analysis, the most recent 400 valid data entries are retrieved from the buffer (or all entries if there are fewer than 400) to form an analysis window.

[0048] Feature calculation (reference) Figure 4 ): Iterate through each SensorData object in the window and calculate the instantaneous composite acceleration a = sqrt(ax² + ay² + az²) and the instantaneous composite angular velocity g = sqrt(gx² + gy² + gz²).

[0049] Find the maximum value of 'a' in the entire window as peakAcceleration, and calculate the average of all 'a' values ​​as avgAcceleration.

[0050] Find the maximum value of g in the entire window as maxGyro, and calculate the average value of all g as avgGyro (optional).

[0051] Calculate the peak acceleration ratio: peakRatio = peakAcceleration / (avgAcceleration + 0.1) (add a small constant to prevent division by zero).

[0052] Estimating Swing Speed Based on Empirical Formula (where k is a calibration coefficient, such as 3.6), Action Classification: Compare the calculated features such as peakAcceleration, maxGyro, peakRatio, avgAcceleration, etc. with the predefined threshold rule set in Example 1 one by one. If all conditions of a certain type of action are met, it is determined as that type of action. If none of them are met, it is marked as "other" or "unknown".

[0053] Quality Assessment: According to the recognized action type, query the corresponding speed quality threshold table for this type. For example, for "smashing", if speed_kmh > 100, it is "excellent"; if 80 < speed_kmh ≤ 100, it is "good"; if speed_kmh ≤ 80, it is "average". Bind the evaluation result with the action type, speed value, etc.

[0054] Package the current action recognition result (sensor ID, student ID, timestamp, action type, speed, quality, etc.) into a MotionAnalysisResult object.

[0055] Trigger the cleaning operation of the sliding window. For example, retain the latest 100 pieces of data in the buffer as the context for the next analysis, and remove older data to balance historical information and computational overhead.

[0056] Submit the MotionAnalysisResult object and the relevant raw data (according to the sampling strategy) to the asynchronous storage task queue.

[0057] Example 3: Implementation of Intelligent Data Sampling Strategy As Figure 3 shown, the system performs sampling decisions on each parsed SensorData object. The logic is as follows (using Java pseudocode example): java private int ordinaryDataCounter = 0; / / Ordinary data counter public boolean shouldSaveData(SensorData data) { / / Rule 1: Key frame data must be saved (such as the start / end points of actions marked by specific algorithms) if (data.isKeyFrame()) {[[ID=3�]] return true; } / / Rule 2: If the data point is associated with an identified action result, it must be stored. / / (Assuming that action IDs or types can be associated with the data object) if (data.getAssociatedMotionId() != null ||data.getDetectedMotionType() != null) { return true; } / / Rule 3: For ordinary data, store 1 record for every 10 records (10% sampling rate) ordinaryDataCounter++; if (ordinaryDataCounter % 10 == 0) { return true; } / / Other cases are not saved return false;} This strategy ensures that all action-related critical information is fully preserved, while significantly compressing the majority of continuous static or low-value dynamic data, thereby significantly reducing storage costs and processing load.

[0058] Example 4: Database Design and Performance Optimization The main data tables of the system are designed as follows: The sensor_data table stores the raw data after sampling and has an index (sensor_id, timestamp) to speed up queries by device and time.

[0059] The `motion_analysis_results` table stores the results of each action recognition, and has indexes `(sensor_id, analysis_time)` and `(student_id, analysis_time)`.

[0060] Performance optimization measures include: Asynchronous batch write: Use the @Async annotation or message queue to put database write operations into a separate thread pool, collect multiple records (such as up to 50 records or wait 100 milliseconds) and then insert them in batches at once.

[0061] Connection pool configuration: Configure database connection pool parameters appropriately (e.g., minimum number of connections 10, maximum number of connections 50) to avoid connection bottlenecks.

[0062] Scheduled cleanup task: Automatically delete old data older than 3 days from the sensor_data and motion_analysis_results tables every day via SpringScheduler or database events.

[0063] Example 5: Test Results In the prototype system testing, 10 sensors were used to simulate a concurrent training scenario, and the following results were obtained: Recognition accuracy: Tests on four intermediate-level athletes, totaling 2,000 standard movements, showed an accuracy rate of 88% for smash, 85% for flat drive, 82% for high clear, and 80% for lob, with an average accuracy rate of 84%.

[0064] System latency: From the arrival of sensor data packets at the server to the updating of the action result on the front-end interface, the average latency is within 800ms, with the core algorithm processing latency being less than 200ms.

[0065] Storage optimization: After enabling intelligent sampling, the daily data storage volume is approximately 12% of the total storage, effectively controlling storage growth.

[0066] System load: Under the pressure of 10 sensors concurrently transmitting 100Hz data streams, the server's average CPU utilization is less than 40%, memory usage is stable, and the system runs stably.

[0067] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart badminton motion analysis system based on motion sensors, characterized in that, The method comprises the following steps: a sensor data collection module for receiving data streams transmitted by motion sensors attached to a badminton racket in real time through a wireless network protocol; a real-time data processing module that maintains a circular buffer for each sensor using a sliding window mechanism to temporarily store continuous motion data; an action recognition analysis engine for extracting multi-dimensional motion features from the data window of the circular buffer and identifying the type of badminton racket action and evaluating its execution quality according to a predefined feature threshold rule set; a data storage management module that uses an intelligent sampling strategy to save key data related to action recognition in its entirety and save continuous sampling data at a reduced scale; a data visualization module for real-time display of action recognition results, quality evaluation, and historical data analysis charts.

2. The motion sensor based smart badminton action analysis system as claimed in claim 1, wherein, In the sliding window mechanism, the window capacity corresponds to N data points. When the data volume in the window reaches M and meets a specific periodic trigger condition, the action recognition analysis engine is triggered to analyze the latest N or all data points in the window. M and N are positive integers, and M≤N.

3. The motion sensor based smart badminton action analysis system as claimed in claim 1 or 2, wherein, The multi-dimensional motion features include: peak and average values of synthetic acceleration, peak value of synthetic angular velocity, and acceleration peak value ratio. The predefined feature threshold rule set corresponds to multiple badminton action types, and the rule for each type is defined by the threshold range of one or more features in the multi-dimensional motion features.

4. The motion sensor based smart badminton action analysis system as claimed in claim 3, wherein, The action types at least include smash, flat shot, high shot, and drop shot. Among them: The rules for identifying a smash include: the peak value of synthetic acceleration is greater than the first acceleration threshold, the peak value of synthetic angular velocity is greater than the first angular velocity threshold, and the acceleration peak value ratio is greater than the first ratio threshold; The rules for identifying a flat shot include: the peak value of synthetic acceleration is greater than the second acceleration threshold, the peak value of synthetic angular velocity is greater than the second angular velocity threshold, and the acceleration peak value ratio is greater than the second ratio threshold; The rules for identifying a high shot include: the peak value of synthetic acceleration is greater than the third acceleration threshold, the average value of synthetic acceleration is greater than the third average acceleration threshold, and the acceleration peak value ratio is within the first ratio interval; The rules for identifying a drop shot include: the peak value of synthetic acceleration is greater than the fourth acceleration threshold, and the peak value of synthetic angular velocity is less than the third angular velocity threshold.

5. The motion sensor based smart badminton action analysis system as claimed in claim 1, wherein, The intelligent sampling strategy includes: Determine whether the current data is pre-labeled key frame data or data associated with action recognition results. If so, save it; If not, classify it as ordinary data and save it at a fixed sampling rate K, where 0 < K < 1.

6. The motion sensor based smart badminton action analysis system as claimed in claim 1, wherein, It also includes a data cleaning module for automatically deleting historical data in the storage module that exceeds the specified time limit according to the pre-set retention strategy.

7. The motion sensor based smart badminton action analysis system as claimed in claim 1, wherein, The real-time data processing module and the data storage management module use an asynchronous processing architecture. The sensor data collection module puts data into a buffer queue, and independent asynchronous task threads perform data processing and storage operations.

8. The motion sensor based smart badminton action analysis system as claimed in claim 1, wherein, The quality evaluation function of the action recognition analysis engine compares the identified action type and its estimated speed value with the pre-set speed quality grading standard of the type of action to output the quality level.

9. The motion sensor based smart badminton action analysis system as claimed in claim 1, wherein, Also include intelligent analysis module, for based on user long-term accumulated action recognition result data, through artificial intelligence model carries out analysis, generates individualized training report and improvement suggestion.

10. A smart badminton action analysis method based on a motion sensor, characterized by, Comprise the following steps: Real-time receive data stream from badminton racket motion sensor, and parse into structured motion data; Add the parsed motion data to the sliding window buffer corresponding to the sensor; When the trigger condition is met, obtain a continuous motion data segment of a specified length from the sliding window buffer; Calculate the multi-dimensional motion feature value of the data segment; Match the calculated feature value with the pre-stored action recognition rule library to determine the type of badminton action corresponding to the current data segment; According to the action type and related feature value, evaluate the execution quality of the action; According to the preset sampling strategy, save part or all of the original motion data and the action recognition and quality evaluation results; Output and display the action type and quality evaluation results.

Images