Intelligent motion data recording method

Abstract

The invention provides a motion data intelligent recording method and a system 1 corresponding to the motion data intelligent recording method. The motion data intelligent recording method comprises the following steps: S1, carrying out time synchronization and denoising preprocessing on original motion data from a multi-source sensor; s2, dividing the preprocessed multi-dimensional motion data into a plurality of overlapped sliding windows according to a time sequence, and caching the sliding windows in a memory; s3, performing multi-dimensional feature judgment on data points except the head and the tail in each sliding window according to a preset priority in sequence, and once any dimension judges that the current point is a feature point, marking the point and terminating judgment of other dimensions; in practical application, through a sliding window structure, multi-dimensional priority judgment, a dynamic threshold mechanism and a forced endpoint retention strategy, efficient, real-time and high-fidelity motion data thinning is realized on resource-constrained equipment, and various technical indexes set for the purpose of the invention are completely met.

Description

Technical Field

[0001] This invention relates to the field of sports data processing technology, and in particular to a method and system for intelligent recording and feature preservation of sports data, which is applicable to devices with powerful computing power and abundant resources, such as mobile phones and servers, as well as portable sports devices with general computing power and limited resources. Background Technology

[0002] With the development of the sports and health industry, devices such as sports watches, smart bracelets, and cycling computers can collect multi-dimensional sports data such as GPS track, heart rate, speed, cadence, and power.

[0003] However, high-frequency data acquisition results in a massive amount of data, which not only occupies storage space but also increases transmission and analysis costs.

[0004] Traditional data thinning methods (such as equal-interval sampling and fixed-threshold filtering) do not consider the characteristic correlation of motion data and are prone to losing key information (such as power peaks during sprints and inflection points of curve trajectories), affecting the accuracy of motion analysis.

[0005] Most current motion data thinning techniques suffer from the following drawbacks: Single-dimensional thinning ignores the correlation between data, such as not retaining the corresponding cadence / power data when deleting speed abrupt changes; Fixed thresholds cannot be adapted to different exercise intensities (e.g., jogging and sprinting require different thinning strategies). The algorithm complexity is too high (such as high-order filtering and global optimization), making it unable to run on low-computing-power devices such as sports watches; Feature extraction relies on global data (such as a complete trajectory), making it impossible to process data while it is being collected. Failure to consider device resource limitations (such as CPU, storage, and power consumption) can easily lead to device lag. To address the aforementioned shortcomings, this invention proposes an intelligent data thinning solution that balances data compression and feature preservation, and can run smoothly on both high-computing-power devices such as mobile phones / servers and low-computing-power devices such as fitness trackers / watches. Summary of the Invention

[0006] The problem with the prior art that this application addresses is: This invention addresses the technical shortcomings of existing motion data thinning methods, such as insufficient feature retention, low compression efficiency, high algorithm complexity, inability to process in real time, and difficulty in adapting to resource-constrained devices. It proposes an intelligent motion data recording method and system. The method introduces a multi-dimensional collaborative judgment mechanism, a dynamic threshold adjustment strategy, a sliding window caching structure, and a priority processing flow. This significantly reduces data storage and transmission overhead while efficiently preserving key semantic feature information during motion. It is applicable to a variety of hardware platforms, from low-power wearable devices to high-performance servers.

[0007] The solution to the technical problem of this invention is: A method for intelligent recording of motion data is provided, including the following steps: S1: Perform time synchronization and noise reduction preprocessing on the raw motion data from multiple sensors; S2: Divide the preprocessed multidimensional motion data into multiple overlapping sliding windows in chronological order and cache them in memory; S3: For each data point in the sliding window except for the first and last points, perform multi-dimensional feature judgment in sequence according to the preset priority. Once any dimension determines that the current point is a feature point, mark the point and terminate the judgment of the remaining dimensions. S4: Force the retention of the first and last groups of data points in each sliding window as feature points; S5: Only write all data marked as feature points to local storage or upload to a remote server, and discard the rest of the original data.

[0008] Preferably, in step S1, the time synchronization process uses the timestamp output by the GPS module as a reference to interpolate and align the heart rate, speed, cadence, power, and altitude data. The noise reduction process uses a first-order low-pass filter with a cutoff frequency set to 0.5Hz to 2Hz.

[0009] Preferably, in step S2, the sliding window consists of n consecutive sets of time-aligned multidimensional data, each set of data containing values ​​of six dimensions: GPS latitude and longitude, heart rate, speed, cadence, power, and altitude at the same time. There is an overlap of one data point between adjacent sliding windows; The sliding window length n ranges from 5 to 30 and is dynamically adjusted based on the device's current available memory capacity, CPU frequency, and remaining battery power.

[0010] Preferably, when the device is in a low-power mode or the remaining memory is below a preset threshold, the sliding window length n is adjusted to 5 to 10; when the device has sufficient resources, the sliding window length n is extended to 15 to 30.

[0011] Preferably, in step S3, the multi-dimensional feature judgment is performed in the following priority order: altitude dimension → physiological and exercise parameter dimension → GPS trajectory dimension; wherein the physiological and exercise parameter dimension includes heart rate, speed, cadence and power.

[0012] Preferably, in determining the altitude dimension, the absolute value of the altitude difference between the current point and the previous point, ΔH, is calculated as ΔH = |Hᵢ−H. i-1 If ΔH > 2 meters, then the i-th set of data is marked as a feature point.

[0013] Preferably, in the determination of physiological and exercise parameters, for any parameter Y∈{heart rate, speed, cadence, power}, the current point P1(Xᵢ,Yᵢ) and its adjacent points P0(Xᵢ,Xᵢ) are taken. i-1 ,Y i-1 P2(X) i+1 ,Y i+1 Construct a straight line segment P0P2 and calculate the perpendicular distance D from point P1 to the line. If D exceeds the dynamic threshold T_Y, then the i-th group of data is marked as a feature point; the dynamic threshold T_Y is adaptively adjusted according to the current motion intensity.

[0014] Preferably, when the speed V > 20 km / h or the power P > 300 W, T_Y = T0 × 1.5; when the speed V ≤ 10 km / h and the power P ≤ 150 W, T_Y = T0 × 0.6. In the intermediate state, T_Y is linearly interpolated between 0.6T0 and 1.5T0; where T0 is the baseline threshold, ranging from 0.8 to 1.5, with the unit depending on the parameter.

[0015] Preferably, in the determination of GPS trajectory dimension, P0 is set as the previous confirmed feature point, P1 is the previous confirmed feature point, and P2 is the current point to be determined; the true geographical vertical distance d_geo from P2 to line segment P0P1 is calculated. If d_geo > T_gps, then P2 is marked as a new feature point, and its previous set of original data points are simultaneously marked as inflection point features; T_gps is adaptively adjusted according to the local trajectory curvature κ.

[0016] Preferably, the curvature κ passes through three points P. a-1 P a P a+1 The calculation yields κ = |θ2−θ1| / Δt, where θ1 and θ2 are the azimuth angles of two adjacent displacement vectors, and Δt is the time interval. When κ > 0.15 rad / s, T_gps = 3 meters; when κ < 0.03 rad / s, T_gps = 8 meters. In the intermediate curvature interval, T_gps is linearly interpolated within the range of 3 meters to 8 meters.

[0017] The technical effects achieved by this application in solving the technical problem are as follows: Compared with existing technologies, the present invention provides an intelligent motion data recording method and a corresponding system 1, which employs the following steps: S1: performing time synchronization and noise reduction preprocessing on raw motion data from multiple sources; S2: dividing the preprocessed multidimensional motion data into multiple overlapping sliding windows in chronological order and caching them in memory; S3: performing multidimensional feature judgment on data points within each sliding window (excluding the first and last points) according to a preset priority; once any dimension determines the current point as a feature point, that point is marked and the judgment of other dimensions is terminated; S4: forcibly retaining the first and last groups of data points in each sliding window as feature points; S5: only writing all data marked as feature points to local storage or uploading to a remote server, discarding the remaining raw data. In practical applications, through a sliding window structure (… Figure 3 ), multi-dimensional priority judgment ( Figure 4 – Figure 9 The dynamic threshold mechanism and forced endpoint retention strategy enable efficient, real-time, and high-fidelity motion data thinning on resource-constrained devices, fully meeting the technical specifications set for the purpose of this invention. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the extraction of heart rate / speed / cadence / power feature values ​​in an intelligent motion data recording method according to the present invention.

[0019] Figure 2 This is a schematic diagram illustrating the GPS trajectory feature value extraction of a smart motion data recording method according to the present invention.

[0020] Figure 3 : This is a flowchart of the window feature value extraction process for an intelligent motion data recording method according to the present invention.

[0021] Figure 4 This is a comparison image showing the elevation feature value extraction before and after the extraction of the motion data intelligent recording method of the present invention.

[0022] Figure 5 This is a comparison image showing the extraction of heart rate feature values ​​before and after using the intelligent recording method for sports data according to the present invention.

[0023] Figure 6 This is a comparison image showing the speed feature value extraction before and after in the intelligent motion data recording method of the present invention.

[0024] Figure 7 This is a comparison image showing the extraction of cadence feature values ​​before and after using the intelligent recording method for motion data of the present invention.

[0025] Figure 8This is a comparison image showing the power feature value extraction before and after the intelligent recording method for motion data of the present invention.

[0026] Figure 9 This is a comparison image showing the GPS trajectory feature value extraction before and after the present invention's intelligent motion data recording method. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0028] It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the invention.

[0029] 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.

[0030] It should be noted that when a component is said to be "fixed" to another component, it can be directly on the other component or it can be in a middle component. When a component is said to be "connected" to another component, it can be directly connected to the other component or it may be in a middle component.

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0032] The terminology used herein in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0033] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0034] Please see Figures 1 to 9 The preferred embodiments of the present invention will be described in detail below. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the scope of protection of the present invention. In this embodiment, a smart cycling computer that supports multi-dimensional motion data acquisition is used as the hardware carrier. It has a built-in GPS module, an optical heart rate sensor, a speed / cadence magnetic sensor, a power meter and a barometric altimeter. It has Bluetooth communication capability and local flash memory. The main control chip is based on the ARM Cortex-M4 architecture with a main frequency of 80MHz and a RAM capacity of 64KB, which is suitable for resource-constrained embedded scenarios. After the system starts, step S1 is executed first: time synchronization and noise reduction preprocessing are performed on the raw motion data from multiple sensors; such as... Figure 3 As shown, the system uses the timestamp output by the GPS module as a reference (e.g., UTC time) to perform linear interpolation and alignment on data such as heart rate, speed, cadence, power, and altitude collected by other sensors, ensuring that each set of multi-dimensional data has a strict correspondence on a unified time axis. Subsequently, a first-order low-pass filter is applied to each dimension of data, with a cutoff frequency set to 1.2Hz, to filter out high-frequency noise caused by sensor jitter or environmental interference, while retaining truly existing abrupt changes. For example, when a user suddenly accelerates, the power data will jump from 150W to 400W within 0.5 seconds; this abrupt change is fully preserved because its frequency is below 1.2Hz. Meanwhile, instantaneous heart rate jumps (>5Hz) caused by road bumps are effectively suppressed. This process is completed by the time synchronization and preprocessing module, the implementation logic of which can be found in [reference needed]. Figure 3 The "data preprocessing" sub-process; Proceed to step S2: Divide the preprocessed multidimensional motion data into multiple overlapping sliding windows in chronological order and cache them; in this embodiment, the initial window length n is set to 10 (i.e., each window contains 10 sets of data sampled once per second, covering 10 seconds); Figure 3 As shown, the sliding window caching module maintains a circular buffer. After the current window finishes processing, the first set of data in the next window reuses the last set of data from the previous window, forming a continuous overlapping structure. The window length n is not fixed, but is dynamically adjusted by the resource adaptation controller: when the battery voltage is detected to be lower than 3.4V or the available memory is lower than 8KB, n is automatically reduced to 6; when the device is powered by USB and the memory margin is greater than 20KB, n is expanded to 20. This mechanism ensures that real-time processing capability can be maintained under different power consumption and storage conditions. Then, step S3 is executed: for the 2nd to n-1th groups of data (i.e., non-first and last points) within each sliding window, multi-dimensional feature judgments are performed sequentially according to priority; such as... Figure 3 As shown in the flowchart, the judgment order is: altitude → heart rate / speed / cadence / power → GPS track; taking the i-th data set in the window (2≤i≤9) as an example, the altitude dimension judgment is performed first; as follows... Figure 4 As shown, the system calculates the difference between the current elevation Hᵢ and the previous elevation Hᵢ₋1, ΔH=|Hᵢ−Hi-1|. If ΔH>2 meters (e.g., a sudden increase from 100 meters to 103 meters), the system immediately marks this set of data as a feature point and skips all subsequent dimension judgments. This threshold is based on statistical analysis of mountain biking test data and can effectively identify the start of an uphill climb or the turning point of a downhill climb, avoiding false triggering due to air pressure fluctuations (usually <1 meter). If altitude does not trigger the specified characteristics, the assessment then proceeds to the physiological and exercise parameter dimensions; for example, heart rate... Figure 5As shown, three points are selected: P0(Xi-1,HRi-1), P1(Xᵢ,HRᵢ), and P2(Xi+1,HRi+1), where X is the timestamp (in seconds) and HR is the heart rate value (in bpm). The system calculates the vertical distance D from P1 to the line P0P2. Assuming the current speed is 25km / h and the power is 320W, it is determined to be a high-intensity state, and the dynamic threshold T_HR = 1.2 × 1.5 = 1.8 bpm. If the calculated D = 2.1 bpm > 1.8 bpm, it is marked as a feature point. Similar logic applies to speed ( Figure 6 ), cadence ( Figure 7 ) and power ( Figure 8 The baseline thresholds T0 for each parameter are: speed 1.0km / h, cadence 3rpm, and power 15W, all calibrated based on actual measurement data. If none of the above dimensions are triggered, then the GPS trajectory dimension is determined; Figure 2 and Figure 9 As shown, let P0 be the "previous confirmed feature point", P1 be the "previous feature point", and P2 be the current point to be determined. The system uses the Haversine formula to convert latitude and longitude into spherical distance, and then uses spherical trigonometry to calculate the true geographic vertical distance d_geo (unit: meters) from P2 to the great circle arc P0P1. At the same time, the system calculates the local curvature κ based on the three nearest points: take P a ₋1、P a P a ₊1, calculate the displacement vector azimuth angles θ1 and θ2 respectively, κ=|θ2−θ1| / Δt; if κ=0.18rad / s (sharp bend), then T_gps=3 meters; if d_geo=3.5 meters>3 meters, then mark P2 as a new feature point, and simultaneously mark its previous set of original points as inflection point features to accurately capture the bend entrance; this mechanism in Figure 9 The trajectory comparison clearly shows that the original dense trajectory (dashed line) retained all the key inflection points (solid circles) after thinning, while the straight line segments were greatly simplified. In step S4, regardless of whether the feature conditions are met, the first and nth groups of data for each window are forcibly retained; for example... Figure 3 As shown, this design ensures that adjacent windows share endpoints when splicing, avoiding trajectory breakage; for example, if the end point of the previous window (i.e. the beginning point of the next window) happens to be on a flat road section, although it is not judged as a feature by any dimension, it still becomes a connection point due to forced preservation, ensuring the continuity of the entire path. Finally, step S5 is executed: only all labeled feature point data (including timestamps and all six-dimensional values) are written to flash memory; unlabeled data is released from the circular buffer immediately after window processing is complete; taking a 1-hour cycling data segment as an example, the original sampling rate is 1Hz, with a total of 3600 data sets; after processing by this method, approximately 600 feature points are retained, achieving a compression rate of 83.3%, while Figures 4 to 9 The comparison showed that all altitude changes, heart rate inflections, power peaks, and trajectory inflection points were completely preserved, with a core feature retention rate of over 95%. Throughout the system's operation, the resource adaptation controller continuously monitors CPU load, memory usage, and battery status. When it detects that three consecutive windows take more than 100ms to process (indicating computing power constraints), the system automatically reduces the window length n from 15 to 8 and replaces floating-point operations in GPS distance calculation with fixed-point approximations (such as using a lookup table instead of square root operations), thereby reducing computational overhead. Conversely, in mobile app scenarios (high-computing-power platforms), n can be set to 30, and more refined curvature calculations can be enabled to further improve feature extraction accuracy. In summary, this embodiment utilizes a sliding window structure ( Figure 3 ), multi-dimensional priority judgment ( Figure 4 – Figure 9 The dynamic threshold mechanism and forced endpoint retention strategy enable efficient, real-time, and high-fidelity motion data thinning on resource-constrained devices, fully meeting the technical specifications set for the purpose of this invention.

[0035] In some other embodiments, this invention discloses an intelligent method and system for recording motion data, relating to the field of motion data processing technology. This method targets multi-dimensional motion data such as GPS trajectory, heart rate, speed, cadence, power, and altitude. It employs a thinning strategy that combines single-dimensional and multi-dimensional collaborative extraction, significantly reducing data storage while maximizing the preservation of key features of the original data (such as trajectory inflection points, extreme points, trend abrupt changes, and multi-dimensional correlation features). This invention solves the problems of traditional thinning methods, such as the easy loss of key motion features and high computational cost. It can achieve a data compression rate of 60%-85% and a core feature retention rate of ≥95%, making it suitable for efficient data recording in devices such as sports watches, cycling computers, and fitness apps.

[0036] The core objective of this invention is to provide a method and system for motion data thinning, achieving the following goals: 1. Feature Preservation: The retention rate of core feature points (inflection points, extreme values) is ≥95%; 2. Compression ratio: Data compression ratio 60%-85%; 3. Resource adaptation: It can adapt to both high-computing-power and low-computing-power devices; 4. Dynamic adaptation: Adjust algorithm parameters according to exercise intensity and equipment status.

[0037] To achieve the above objectives, the present invention adopts the following technical solution: I. Data Preprocessing: Time synchronization: Based on GPS timestamps, calibrate the timestamps of GPS latitude and longitude, heart rate, speed, cadence, power, altitude, and other data. Noise reduction: Simple smoothing filtering is applied to GPS latitude and longitude, heart rate, speed, cadence, power, altitude and other data to preserve real abrupt changes in signal.

[0038] II. Collaborative Feature Extraction Based on Single-Dimensional and Multi-Dimensional Dimensions: Window data caching: Cache one window of data every n seconds (one set of data per second, n sets of data per window). Except for the first window of data, the first set of data for subsequent windows is the last set of data of the previous window (the size of n can be selected according to device resources. The larger n is, the larger the cache required and the greater the amount of computation). Window feature extraction: The first and last groups of data in the window are fixed as feature values. The data from groups 2 to n-1 in the window are processed iteratively. Each group is processed in the following order: once a data item is determined to be a feature value, this group of data is marked as a feature value, and feature value extraction for subsequent data items is no longer performed; the process proceeds directly to the next group of data. Altitude: Compare the current altitude with the previous altitude. If the difference is greater than 2 meters, it is marked as a feature value. Heart rate / speed / cadence / power: With timestamp as the X-axis and heart rate / speed / cadence / power data as the Y-axis, three points are selected: P0(Xi-1, Yi-1), P1(Xi, Yi), and P2(Xi+1, Yi+1). P0 represents the previous set of data, P1 represents the current set of data, and P2 represents the next set of data. When the vertical distance from point P1 to the line connecting P0 and P2 exceeds a threshold (D), P1 is marked as a feature value. The distance threshold (D) is dynamically adjusted based on exercise intensity. GPS trajectory: With longitude as the X-axis and latitude as the Y-axis, take three GPS points P0(X0, Y0), P1(X1, Y1), and P2(Xi,Yi). P0 is the previous feature point, P1 is the previous feature point, and P2 is the current point. When the actual geographical vertical distance from point P2 to the straight line from P0 to P1 exceeds the threshold (D), P2 is marked as a feature value. The previous set of data for P2 is the inflection point and is also marked as a feature value. The distance threshold (D) is dynamically adjusted according to the trajectory curvature.

[0039] Compared with existing technologies, the present invention provides an intelligent motion data recording method and a corresponding system 1, which employs the following steps: S1: performing time synchronization and noise reduction preprocessing on raw motion data from multiple sources; S2: dividing the preprocessed multidimensional motion data into multiple overlapping sliding windows in chronological order and caching them in memory; S3: performing multidimensional feature judgment on data points within each sliding window (excluding the first and last points) according to a preset priority; once any dimension determines the current point as a feature point, that point is marked and the judgment of other dimensions is terminated; S4: forcibly retaining the first and last groups of data points in each sliding window as feature points; S5: only writing all data marked as feature points to local storage or uploading to a remote server, discarding the remaining raw data. In practical applications, through a sliding window structure (… Figure 3 ), multi-dimensional priority judgment ( Figure 4 – Figure 9 The dynamic threshold mechanism and forced endpoint retention strategy enable efficient, real-time, and high-fidelity motion data thinning on resource-constrained devices, fully meeting the technical specifications set for the purpose of this invention.

[0040] The embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for intelligent recording of motion data, characterized in that: Includes the following steps, S1: Perform time synchronization and noise reduction preprocessing on the raw motion data from multiple sensors; S2: Divide the preprocessed multidimensional motion data into multiple overlapping sliding windows in chronological order and cache them in memory; S3: For each data point in the sliding window except for the first and last points, perform multi-dimensional feature judgment in sequence according to the preset priority. Once any dimension determines that the current point is a feature point, mark the point and terminate the judgment of the remaining dimensions. S4: Force the retention of the first and last groups of data points in each sliding window as feature points; S5: Only write all data marked as feature points to local storage or upload to a remote server, and discard the rest of the original data.

2. The intelligent motion data recording method as described in claim 1, characterized in that: In step S1, the time synchronization process uses the timestamp output by the GPS module as a reference to interpolate and align the heart rate, speed, cadence, power, and altitude data. The noise reduction process uses a first-order low-pass filter with a cutoff frequency set to 0.5Hz to 2Hz.

3. The method for intelligent recording of motion data as described in claim 2, characterized in that: In step S2, the sliding window consists of n consecutive sets of time-aligned multidimensional data, each set of data containing the values ​​of six dimensions: GPS latitude and longitude, heart rate, speed, cadence, power, and altitude at the same time. There is an overlap of one data point between adjacent sliding windows; The sliding window length n ranges from 5 to 30 and is dynamically adjusted based on the device's current available memory capacity, CPU frequency, and remaining battery power.

4. The method for intelligent recording of motion data as described in claim 3, characterized in that: When the device is in low-power mode or the remaining memory is below a preset threshold, the sliding window length n is adjusted to 5 to 10; when the device has sufficient resources, the sliding window length n is extended to 15 to 30.

5. The method for intelligent recording of motion data as described in claim 1, characterized in that: In step S3, the multi-dimensional feature judgment is performed in the following priority order: altitude dimension → physiological and exercise parameter dimension → GPS trajectory dimension; wherein the physiological and exercise parameter dimension includes heart rate, speed, cadence and power.

6. The method for intelligent recording of motion data as described in claim 5, characterized in that: In determining altitude, the absolute value of the altitude difference between the current point and the previous point is calculated as ΔH = |Hᵢ−H. i-1 If ΔH > 2 meters, then the i-th set of data is marked as a feature point.

7. The intelligent recording method for motion data as described in claim 5, characterized in that: In the determination of physiological and exercise parameters, for any parameter Y∈{heart rate, speed, cadence, power}, take the current point P1(Xᵢ,Yᵢ) and its adjacent points P0(Xᵢ,Xᵢ,Yᵢ) before and after it. i-1 ,Y i-1 P2(X) i+1 ,Y i+1 Construct a straight line segment P0P2 and calculate the perpendicular distance D from point P1 to the line. If D exceeds the dynamic threshold T_Y, then the i-th group of data is marked as a feature point; the dynamic threshold T_Y is adaptively adjusted according to the current motion intensity.

8. The intelligent recording method for motion data as described in claim 7, characterized in that: When speed V > 20 km / h or power P > 300 W, T_Y = T0 × 1.5; when speed V ≤ 10 km / h and power P ≤ 150 W, T_Y = T0 × 0.

6. In the intermediate state, T_Y is linearly interpolated between 0.6T0 and 1.5T0; where T0 is the baseline threshold, ranging from 0.8 to 1.5, with the unit depending on the parameter.

9. The intelligent recording method for motion data as described in claim 5, characterized in that: In the determination of GPS trajectory dimension, let P0 be the previous confirmed feature point, P1 be the previous confirmed feature point, and P2 be the current point to be determined; calculate the true geographic vertical distance d_geo from P2 to line segment P0P1. If d_geo > T_gps, then mark P2 as a new feature point and simultaneously mark its previous set of original data points as inflection point features; T_gps is adaptively adjusted according to the local trajectory curvature κ.

10. The method for intelligent recording of motion data as described in claim 9, characterized in that: The curvature κ passes through three points P a-1 P a P a+1 The calculation yields κ = |θ2−θ1| / Δt, where θ1 and θ2 are the azimuth angles of two adjacent displacement vectors, and Δt is the time interval. When κ > 0.15 rad / s, T_gps = 3 meters; when κ < 0.03 rad / s, T_gps = 8 meters. In the intermediate curvature interval, T_gps is linearly interpolated within the range of 3 meters to 8 meters.

Images