Human motion posture data acquisition system based on convolutional neural network

Abstract

The invention discloses a human motion posture data acquisition system based on a convolutional neural network, and belongs to the technical field of human motion posture data processing. The system comprises a signal error correction module, a motion influence correction module and a data integration module, the signal error correction module judges whether to perform GNSS data dynamic correction or not according to the signal error parameters; the motion influence correction module judges whether IMU data dynamic correction and GNSS speed data dynamic correction are carried out or not according to motion influence parameters of all joints in all motion periods in the data acquisition process of the IMU sensor; and the data integration module performs window time dynamic correction according to the motion characteristics of the IMU data of each motion period, integrates the IMU data after window time correction judgment and the GNSS data after motion influence quantification judgment to obtain human motion posture data, and realizes high-precision acquisition of the human motion posture data.

Description

Technical Field

[0001] The present invention relates to the technical field of human motion posture data processing, and in particular to a human motion posture data acquisition system based on a convolutional neural network. Background Art

[0002] Through precise data analysis and modeling, human motion posture data acquisition systems can evaluate athletes' performance, optimize training programs, and help develop personalized treatment plans for rehabilitation patients. With the advancement of technology, the accuracy of sensors continues to improve, and the application scope of the system is gradually expanding, providing important support for improving human health and exercise efficiency.

[0003] The existing human motion posture data acquisition system uses sensors to capture motion data of various parts of the human body in real time, combines computer vision and machine learning algorithms to process and analyze the data, and provides accurate motion trajectory, posture and action evaluation. The system transmits data to the processing platform via wireless network to achieve real-time feedback and motion optimization.

[0004] For example, the invention patent announcement with announcement number CN112287840B discloses a method and system for intelligently collecting motion analysis data, including: using the multimedia processing tool FFmpeg to push the video stream collected by the camera to the background RSTP server in the form of video push streaming; using the human body posture model to identify the human target boundary box and human body movements in the video stream, extracting video frames in a static state from the video stream, extracting the static frames of the observed person frame by frame from the video stream, and dynamically inserting the static frames into the static frame chain in chronological order.

[0005] For example, the invention patent announcement with publication number CN111368667B discloses a data acquisition method, electronic device, and storage medium, including: selecting a target person from a plurality of stored person models, and / or selecting a target scene from a plurality of stored scene models; controlling the movement of the target person in the target scene; and obtaining motion data of the target person at different acquisition times, wherein each motion data includes: an image of the target person and position information of each joint of the target person in the image.

[0006] However, in the process of implementing the technical solutions of the invention in the embodiments of the present application, the present application found that the above technology has at least the following technical problems:

[0007] In the existing technology, when collecting human motion posture data during an athlete's exercise, the IMU (Inertial Measurement Unit) sensor calculates posture and position through integration. When the error in the angular velocity measured by the gyroscope in the IMU is integrated into an angle, the error accumulates linearly over time, causing the sensor noise to be continuously amplified during the integration process. As a result, the error in the human motion posture data obtained by the IMU sensor increases, resulting in low accuracy in collecting human motion posture data during strenuous exercise in outdoor environments. Summary of the Invention

[0008] The embodiments of the present application provide a human motion posture data acquisition system based on a convolutional neural network, thereby solving the problem in the prior art of collecting human motion posture data during an athlete's exercise. Since the IMU sensor calculates the posture and position through integration, the error of the gyroscope in the IMU in measuring the angular velocity accumulates linearly with time when integrated into an angle, causing the sensor noise to be continuously amplified during the integration process. As a result, the error of the human motion posture data acquired by the IMU sensor increases, resulting in low accuracy of the human motion posture data collected during strenuous exercise in an outdoor environment. The system achieves high-precision collection of human motion posture data.

[0009] The embodiment of the present application provides a human motion posture data acquisition system based on a convolutional neural network, comprising: a signal error correction module, a motion influence correction module, a data integration module and a motion posture database; wherein the signal error correction module is used to perform a quantitative determination of the multipath effect according to the signal error parameters in the data acquisition process of the GNSS sensor, obtain a quantitative determination result of the multipath effect, and determine whether to perform dynamic correction of the GNSS data based on the quantitative determination result of the multipath effect, wherein the dynamic correction of the GNSS data means correcting the GNSS data according to the signal error parameters to reduce the influence of the multipath effect on the GNSS data, and the GNSS data is the GNSS position data and GNSS speed data collected by the GNSS sensor; the motion influence correction module is used to perform a quantitative determination of the motion influence according to the motion influence parameters of each joint in each motion cycle during the data acquisition process of the IMU sensor, and obtain a motion influence quantity. The dynamic correction of IMU data means correcting IMU data according to motion influence parameters to reduce the influence of motion state on IMU data. The dynamic correction of GNSS velocity data means correcting GNSS velocity data according to motion influence parameters to reduce the influence of motion state on GNSS velocity data. A data integration module is used to judge whether to perform dynamic correction of window time on IMU data of each motion cycle according to the motion characteristics of IMU data of each motion cycle, and integrate the IMU data after window time correction and the GNSS data after motion influence quantification to obtain human body motion posture data. The dynamic correction of window time means correcting the window time of IMU data of each motion cycle according to the motion characteristics of IMU data of each motion cycle to reduce the influence of window time difference on the efficiency of convolutional neural network feature extraction process.

[0010] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0011] 1. The present invention provides a human motion posture data acquisition system based on a convolutional neural network, thereby effectively correcting signal errors and motion effects, reducing the interference of multipath effects on GNSS data, optimizing the motion state impact of IMU data, and improving data integration accuracy through window time dynamic correction, thereby achieving accurate and efficient human motion posture data acquisition, improving the efficiency of the convolutional neural network feature extraction process, and ensuring the stability and reliability of the system in various motion environments.

[0012] 2. The present invention determines whether to perform dynamic correction of GNSS data based on the quantitative determination results of the multipath effect, thereby taking corresponding processing measures under different signal error conditions, thereby optimizing the stability and reliability of GNSS data and ensuring the high accuracy and stability of the overall system.

[0013] 3. The present invention determines whether to perform dynamic correction of IMU data and GNSS velocity data based on the quantitative determination result of motion impact, thereby taking corresponding processing measures under different motion states, thereby realizing dynamic correction of IMU data and GNSS velocity data, improving the accuracy and reliability of the data, and realizing precise correction under different motion characteristics, ensuring the stability and accuracy of the system in complex motion environments.

[0014] 4. The present invention dynamically corrects the weights of IMU data and GNSS data according to the credibility parameters of IMU data after window time classification and GNSS data after motion impact quantification, thereby dynamically adjusting the contribution of IMU data and GNSS data to state estimation during the unscented Kalman filter integration process, thereby optimizing the accuracy and reliability of the final state estimation and ensuring the high accuracy and stability of human motion posture data. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of the structure of a human motion posture data acquisition system based on a convolutional neural network provided in an embodiment of the present application.

[0016] Figure 2 A mind map of the signal error correction module provided in an embodiment of the present application.

[0017] Figure 3 A mind map of the motion impact correction module provided in an embodiment of the present application.

[0018] Figure 4 This is a flowchart of the algorithm for the unscented Kalman filter provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The embodiment of the present application provides a human motion posture data acquisition system based on a convolutional neural network, which solves the problem in the prior art that when collecting human motion posture data during an athlete's exercise, the IMU sensor calculates the posture and position through integration, and the error of the gyroscope in the IMU in measuring the angular velocity accumulates linearly with time when the error is integrated into an angle, resulting in the sensor noise being continuously amplified during the integration process, and the error of the human motion posture data acquired by the IMU sensor increases. There is a problem of low accuracy of human motion posture data collected during intense exercise in an outdoor environment. By quantitatively judging the influence of multipath effects according to the signal error parameters in the data acquisition process of the GNSS sensor, a quantitative judgment result of the influence of multipath effects is obtained, and whether to perform dynamic correction of GNSS data is determined based on the quantitative judgment result of the influence of multipath effects. The dynamic correction of GNSS data means correcting the GNSS data according to the signal error parameters to reduce the influence of multipath effects on the GNSS data. The GNSS data is GNSS position data and GNSS velocity data collected by the GNSS sensor; according to the signal error parameters of the IMU sensor, the GNSS data is corrected according to the signal error parameters to reduce the influence of multipath effects on the GNSS data. The motion influence parameters of each joint in each motion cycle during the data acquisition process of the sensor are used to quantitatively determine the motion influence, and the motion influence quantitative determination results are obtained. Based on the motion influence quantitative determination results, it is determined whether to perform dynamic correction of IMU data and dynamic correction of GNSS speed data. Dynamic correction of IMU data means correcting the IMU data according to the motion influence parameters to reduce the influence of the motion state on the IMU data. Dynamic correction of GNSS speed data means correcting the GNSS speed data according to the motion influence parameters to reduce the influence of the motion state on the GNSS speed data. According to the motion characteristics of the IMU data of each motion cycle, it is determined whether to perform dynamic correction of the window time of the IMU data of each motion cycle. The IMU data after the window time correction and the GNSS data after the motion influence quantitative determination are integrated to obtain the human motion posture data. Dynamic correction of window time means correcting the window time of the IMU data of each motion cycle according to the motion characteristics of the IMU data of each motion cycle to reduce the influence of the window time difference on the efficiency of the convolutional neural network feature extraction process, thereby realizing the collection of high-precision human motion posture data.

[0020] The technical solution in the embodiments of the present application is to solve the above-mentioned problem of collecting human motion posture data during athlete exercise. Since the IMU sensor calculates posture and position through integration, the error of the gyroscope in the IMU in measuring angular velocity accumulates linearly over time when integrated into an angle, resulting in the sensor noise being continuously amplified during the integration process. The error of the human motion posture data obtained by the IMU sensor increases, resulting in low accuracy of human motion posture data collected during strenuous exercise in outdoor environments. The overall idea is as follows:

[0021] By quantifying the multipath effects in GNSS signals and the motion-affecting parameters of each joint in the IMU sensor's motion cycle, the authors determine whether to dynamically correct the GNSS and IMU data, thereby reducing the impact of multipath and motion on the data. Furthermore, the IMU data is dynamically corrected for window time based on motion characteristics during the information extraction process, minimizing the impact of window time differences on the efficiency of convolutional neural network feature extraction. Ultimately, by integrating the corrected IMU and GNSS data, they achieve high-precision human motion posture data acquisition.

[0022] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0023] like Figure 1As shown, it is a structural diagram of a human motion posture data acquisition system based on a convolutional neural network provided in an embodiment of the present application. The human motion posture data acquisition system based on a convolutional neural network provided in an embodiment of the present application includes: a signal error correction module, a motion impact correction module, a data integration module and a motion posture database; wherein the signal error correction module is used to perform a quantitative determination of the multipath effect according to the signal error parameter in the data acquisition process of the GNSS sensor, obtain a quantitative determination result of the multipath effect, and determine whether to perform GNSS (Global Navigation Satellite System, Global Navigation Satellite System) data dynamic correction, GNSS data dynamic correction means correcting GNSS data according to signal error parameters to reduce the impact of multipath effect on GNSS data, GNSS data is GNSS position data and GNSS velocity data collected by GNSS sensor; motion impact correction module is used to quantitatively determine the motion impact according to the motion impact parameters of each joint in each motion cycle during the data collection process of IMU sensor, obtain the motion impact quantitative determination result, and determine whether to perform IMU data dynamic correction and GNSS velocity data dynamic correction based on the motion impact quantitative determination result. IMU data dynamic correction means correcting IMU data according to motion impact parameters to reduce the impact of motion state on IMU data. The U data is the angular velocity data and acceleration data collected by the IMU sensor. The dynamic correction of the GNSS velocity data means correcting the GNSS velocity data according to the motion influence parameters to reduce the influence of the motion state on the GNSS velocity data. The data integration module is used to determine whether to dynamically correct the window time of the IMU data of each motion cycle according to the motion characteristics of the IMU data of each motion cycle, and integrate the IMU data after the window time correction and the GNSS data after the motion influence quantification judgment to obtain the human body motion posture data. The motion characteristics include low-frequency motion and high-frequency motion. The dynamic correction of the window time means correcting the window time of the IMU data of each motion cycle according to the motion characteristics of the IMU data of each motion cycle to reduce the influence of the window time difference on the efficiency of the convolutional neural network feature extraction process.

[0024] In addition, the motion posture database is used to store relevant data of the human motion posture data acquisition system based on convolutional neural networks, including: critical carrier phase residual, critical pseudorange multipath error, critical signal-to-noise ratio, carrier phase residual contribution, signal error first threshold and signal error second threshold, etc. The data in the motion posture database can be directly queried through public databases of human posture estimation and motion capture such as the MPII (MPII Human Pose Dataset) dataset, and can be obtained through cooperation with corporate departments in the fields of sports competition, robotic automation or medical rehabilitation.

[0025] In this embodiment, the IMU sensor can obtain the athlete's angular velocity and acceleration data through a gyroscope and accelerometer, and combine this with integral calculations to obtain the real-time attitude angles and motion trajectories of various parts of the human body, where the attitude angles include pitch, roll, and yaw. GNSS can obtain the athlete's three-dimensional coordinates and velocity information by receiving satellite signals, but cannot perceive the body's joint movements. The IMU has complementary characteristics to the GNSS. The IMU error is initially very small, but it becomes increasingly larger over time. While the GNSS signal is relatively less accurate in the short term, its accuracy is cyclical. The present invention integrates the GNSS and IMU using an unscented Kalman filter, dynamically assigning weights based on the GNSS and IMU credibility, providing real-time, accurate motion data for teaching and training, helping coaches and athletes conduct detailed motion analysis and optimize training results. At the same time, through low-power wireless communication technology, the device can achieve long-term remote monitoring and data transmission, reducing wiring difficulty and maintenance costs, and improving the accuracy of remote detection of athletes' postures.

[0026] like Figure 2 As shown, it is a mind map of signal error correction provided by an embodiment of the present application, including: real-time monitoring of signal error parameters, and obtaining a signal error index based on the signal error parameter analysis, comparing the signal error index with the first signal error threshold and the second signal error threshold respectively, if the signal error index is less than or equal to the first signal error threshold, then the corresponding multipath effect impact quantification determination result is the first signal error result, and no additional processing is performed; if the signal error index is greater than the first signal error threshold and less than or equal to the second signal error threshold, then the corresponding multipath effect impact quantification determination result is the second signal error result, and the GNSS data is corrected according to the difference between the second signal error threshold and the signal error index; if the signal error index is greater than the second signal error threshold, then the corresponding multipath effect impact quantification determination result is the third signal error result, and it is determined whether the GNSS sensor has turned on signal filtering. If so, the filtering noise parameter is adjusted according to the signal error index, otherwise the signal filtering is turned on.

[0027] Specifically, the influence of multipath effect is quantitatively determined according to the signal error parameters in the data acquisition process of the GNSS sensor, and the steps for obtaining the quantitative determination result of the influence of multipath effect include: obtaining signal error parameter reference data from a preset motion posture database, specifically including: critical carrier phase residual, critical pseudorange multipath error and critical signal-to-noise ratio; performing proportion approaching operation on the signal error parameters and the corresponding signal error parameter reference data respectively, the proportion approaching operation is a ratio operation, and then using the signal error parameter contribution parameter to weight the proportion approaching operation results and couple them to obtain a signal error index, the signal error parameter contribution includes the carrier phase residual contribution, the pseudorange multipath error contribution and the signal-to-noise ratio contribution, and the signal error index represents the signal error parameter contribution. Quantify data on the degree of influence on the accuracy of GNSS data, signal error parameters include carrier phase residual, pseudorange multipath error and signal-to-noise ratio; obtain a first signal error threshold and a second signal error threshold from a preset motion posture database; compare the signal error index with the first signal error threshold and the second signal error threshold respectively; if the signal error index is less than or equal to the first signal error threshold, the corresponding quantitative determination result of the multipath effect influence is recorded as the first signal error result; if the signal error index is greater than the first signal error threshold and less than or equal to the second signal error threshold, the corresponding quantitative determination result of the multipath effect influence is recorded as the second signal error result; if the signal error index is greater than the second signal error threshold, the corresponding quantitative determination result of the multipath effect influence is recorded as the third signal error result.

[0028] The signal error index is obtained as follows: ;

[0029] Where, represents the signal error index, represents the carrier phase residual contribution, represents the pseudorange multipath error contribution, represents the signal-to-noise ratio contribution, It represents the carrier phase residual, which is the difference between the carrier phase observation value and the theoretical carrier phase, reflecting the disturbance of the signal propagation path. The larger the carrier phase residual, the larger the signal error index. The carrier phase observation value is the difference obtained by the GNSS system receiver by comparing the received satellite carrier signal phase with the local carrier signal phase generated inside the receiver. The theoretical carrier phase is the expected phase value obtained based on the geometric distance between the satellite and the receiver, clock error, atmospheric delay and other models in the motion attitude database. Represents the pseudorange multipath error, which is the error component in the pseudorange observation value that comes from the multipath effect and cannot be corrected by the model. The larger the pseudorange multipath error, the larger the signal error index. The calculation formula of the pseudorange multipath error is as follows: Where, represents the pseudorange observation value, is the carrier phase observation value, and These are the frequencies of two different carrier signals emitted by satellites preset in the GNSS system. and These are the two main carrier frequencies preset in the GNSS system, both of which can be directly obtained through the dual-frequency observation data module of the receiver in the GNSS system; represents the critical pseudorange multipath error, It represents the signal-to-noise ratio. The larger the signal-to-noise ratio, the smaller the signal error index. It can be directly obtained from the GNSS receiver. represents the critical signal-to-noise ratio.

[0030] It should be noted that in this algorithm, the carrier phase residual, pseudorange multipath error, and signal-to-noise ratio are interrelated. For example, the carrier phase residual and pseudorange multipath error both come from uncertainties in signal propagation. The carrier phase residual affects the accuracy of phase measurement, while the pseudorange multipath error directly affects the distance measurement. The signal-to-noise ratio is directly related to the quality of the signal. The smaller the signal-to-noise ratio, the larger the carrier phase residual and pseudorange multipath error will be.

[0031] 、 and These are the contributions of the carrier phase residual, pseudorange multipath error, and signal-to-noise ratio preset in the motion attitude database. These values ​​represent the degree to which these factors, respectively, contribute to the signal error index. These values ​​can be directly retrieved from the motion attitude database. These relationships are organized into a "mapping set," or lookup table or corresponding rules. When a signal error parameter is actually monitored, the corresponding mapping set can be used to look up that value and determine its corresponding contribution. This contribution, a number between 0 and 1, represents an assessment of the degree to which the current signal error parameter interferes with GNSS data accuracy. For example, for carrier phase residuals, there is a mapping set between carrier phase residuals and their corresponding contributions. By inputting the value of the carrier phase residual, the corresponding contribution of the carrier phase residual can be obtained. For pseudorange multipath error, there is a mapping set between pseudorange multipath error and its corresponding contribution. By inputting the value of the pseudorange multipath error, the corresponding contribution of the pseudorange multipath error can be obtained. For signal-to-noise ratio, there is a mapping set between signal-to-noise ratio and its corresponding contribution. By inputting the value of the signal-to-noise ratio, the corresponding contribution of the signal-to-noise ratio can be obtained. These mapping relationships can be many-to-one or one-to-one.

[0032] As a further solution, the step of judging whether to perform dynamic correction of GNSS data based on the quantitative determination result of the multipath effect includes: if the quantitative determination result of the multipath effect is the first signal error result, no additional processing is performed; if the quantitative determination result of the multipath effect is the second signal error result, the difference between the second signal error threshold and the signal error index is recorded as the signal error deviation index, and the signal error deviation index is matched with the GNSS data correction value corresponding to each signal error deviation index preset in the motion posture database. In the motion posture database, each signal error deviation index and the GNSS data correction value correspond one to one to form a mapping relationship table, which records each signal error deviation index and its corresponding GNSS S data correction value, these relationships can be one-to-one or many-to-one. When obtaining the GNSS data correction value, it is only necessary to input the signal error deviation index into the mapping relationship table. The motion attitude database can quickly locate and return the GNSS data correction value corresponding to the signal error deviation index. The GNSS data correction value is summed with the GNSS data to obtain the corrected GNSS data; if the multipath effect affects the quantitative judgment result of the signal error third result, a GNSS data abnormality prompt is issued to determine whether signal filtering such as Kalman filtering has been turned on. If so, the filter noise parameter is adjusted according to the difference between the signal error index and the signal error second threshold. Otherwise, the signal filtering is turned on. The filter noise parameters include process noise and observation noise.

[0033] Among them, the step of adjusting the filter noise parameter according to the difference between the signal error index and the second threshold value of the signal error includes: obtaining an empirical coefficient from a preset motion posture database; marking the difference between the signal error index and the second threshold value of the signal error as the signal error deviation amplitude, and matching it with the process noise multiples corresponding to each signal error deviation amplitude preset in the motion posture database, and forming a mapping relationship table in which each signal error deviation amplitude range and the process noise multiple correspond one-to-one in the motion posture database. The table records each signal error deviation amplitude range and its corresponding process noise multiple. These relationships can be one-to-one or many-to-one. When obtaining the process noise multiple, it is only necessary to input the signal error deviation amplitude into the mapping relationship table, and the motion posture database can quickly locate and return the process noise multiple corresponding to the signal error deviation amplitude, and adjust the process noise according to the process noise multiple; adjust the observation noise according to the current observation noise, the empirical coefficient and the signal error deviation amplitude. The specific method is as follows: Where, represents the adjusted observation noise, represents the current observation noise, is the empirical coefficient, and MP is the signal error deviation amplitude.

[0034] In this embodiment, a method for dynamically correcting GNSS data based on the quantified determination of multipath effects effectively addresses data errors caused by multipath effects. By precisely calculating the signal error deviation index and combining it with mapping relationships in a motion posture database, GNSS data can be accurately matched and corrected, improving positioning accuracy. When the third signal error result is determined, filtering parameters are adjusted or signal filtering is enabled to further suppress multipath interference, thereby ensuring efficient and stable operation of the GNSS system in complex environments. This dynamic correction mechanism effectively improves the reliability and accuracy of GNSS data, particularly in scenarios with significant multipath effects.

[0035] like Figure 3 As shown, it is a mind map of the motion impact correction module provided by the embodiment of the present application, which specifically includes: real-time monitoring of the motion impact parameters of each joint in each motion cycle, obtaining a motion impact index based on the motion impact parameter analysis, comparing the motion impact index with the motion impact threshold, if the motion impact index is less than or equal to the motion impact threshold, then the corresponding motion impact quantitative determination result is recorded as the first motion impact result, and the corresponding action feature is marked as a low-frequency action, if the motion impact index is greater than the motion impact threshold, then the corresponding motion impact quantitative determination result is recorded as the second motion impact result, and the corresponding action feature is marked as a high-frequency action, and the IMU data and GNSS speed data are corrected according to the difference between the motion impact index and the motion impact threshold.

[0036] As a further solution, the motion influence parameters of each joint in each motion cycle during the data acquisition process of the IMU sensor are used to quantitatively determine the motion influence, and the steps for obtaining the motion influence quantitative determination result include: obtaining motion influence parameter reference data from a preset motion posture database, specifically including: critical acceleration amplitude, critical angular velocity amplitude and critical average acceleration; performing a proportion approach calculation on the motion influence parameters of each joint in each motion cycle and the corresponding motion influence parameter reference data, and then using the motion influence parameter contribution to weight the proportion approach calculation results and then couple them, and then perform average processing on the coupling processing results to obtain the motion influence index of each motion cycle, and the motion influence parameter The contribution degree includes the acceleration amplitude contribution degree, the angular velocity amplitude contribution degree and the average acceleration contribution degree. The motion impact index represents the quantitative data of the degree of influence of the motion impact parameters on the accuracy of the IMU data. The motion impact parameters include the acceleration amplitude, the angular velocity amplitude and the average acceleration. The motion impact threshold is obtained from the preset motion posture database, and the motion impact index of each motion cycle is compared with the motion impact threshold. If the motion impact index of a certain motion cycle does not exceed the motion impact threshold, the corresponding motion impact quantitative determination result is marked as the first motion impact result. If the motion impact index of a certain motion cycle exceeds the motion impact threshold, the corresponding motion impact quantitative determination result is marked as the second motion impact result.

[0037] The exercise impact index of each exercise cycle is obtained as follows: ;

[0038] Where, represents the motion impact index of the mth motion cycle, represents the acceleration amplitude contribution, represents the contribution of angular velocity amplitude, represents the average acceleration contribution, It represents the acceleration amplitude of the jth joint in the mth motion cycle. The larger the acceleration amplitude, the greater the motion impact index. represents the critical acceleration amplitude, It represents the angular velocity amplitude of the jth joint in the mth motion cycle. The larger the angular velocity amplitude, the greater the motion influence index. represents the critical angular velocity amplitude, It represents the average acceleration of the jth joint in the mth motion cycle. The greater the average acceleration, the greater the motion impact index. represents the critical average acceleration, where m is the number of each motion cycle, m=1,2,3,..., , is the total number of motion cycles, j is the joint number, j=1,2,3,..., , is the total number of joints. Acceleration amplitude, angular velocity amplitude, and average acceleration can all be measured by IMU sensors. These three are interrelated. For example, acceleration amplitude and angular velocity amplitude reflect the force and speed of joint movement. A larger acceleration amplitude is associated with a larger angular velocity amplitude. Acceleration amplitude focuses on the maximum acceleration, while average acceleration is the average acceleration over the entire movement cycle. A larger acceleration amplitude for each joint results in a larger average acceleration. Similarly, a larger angular velocity amplitude for each joint results in a larger average acceleration.

[0039] 、 and These are the contributions of the acceleration amplitude, angular velocity amplitude, and average acceleration preset in the motion posture database, respectively. They represent the numerical values ​​of the degree of influence of the acceleration amplitude, angular velocity amplitude, and average acceleration on the motion influence index, and can be directly obtained from the motion posture database when used. These relationships are organized into a "mapping set," or a lookup table or corresponding rules. When the value of a certain motion influencing parameter is actually monitored, this value can be looked up in the corresponding mapping set, and then the corresponding contribution can be obtained. This contribution is a number between 0 and 1, which represents the evaluation result of the degree of interference of the current motion influencing parameter on the human motion posture data. For example, for acceleration amplitude, there is a mapping set between acceleration amplitude and corresponding contribution degree. Inputting the value of acceleration amplitude will return the corresponding contribution degree. For angular velocity amplitude, there is a mapping set between angular velocity amplitude and corresponding contribution degree. Inputting the value of angular velocity amplitude will return the corresponding contribution degree. For average acceleration, there is a mapping set between average acceleration and corresponding contribution degree. Inputting the value of average acceleration will return the corresponding contribution degree. These mapping relationships can be many-to-one or one-to-one.

[0040] As a further solution, the steps of determining whether to perform dynamic correction of IMU data and dynamic correction of GNSS speed data based on the motion impact quantification determination result include: if the motion impact quantification determination result of a certain motion cycle is the first motion impact result, the motion characteristics of the motion cycle are marked as low-frequency motion, and no additional processing is performed; if the motion impact quantification determination result of a certain motion cycle is the second motion impact result, the motion characteristics of the motion cycle are marked as high-frequency motion, and the difference between the motion impact index of the motion cycle and the motion impact threshold is marked as the motion impact deviation index. The IMU data is dynamically corrected according to the motion impact deviation index, and the GNSS speed data is dynamically corrected, so as to reduce the interference of strenuous motion on the accuracy of IMU data and GNSS speed data.

[0041] Among them, the steps of dynamically correcting the IMU data according to the motion influence deviation index and dynamically correcting the GNSS speed data include: matching the motion influence deviation index with the IMU data correction values ​​corresponding to the motion influence deviation indices preset in the motion attitude database, forming a mapping relationship table in which each motion influence deviation index and each IMU data correction value correspond one to one, and the table records each motion influence deviation index and its corresponding IMU data correction value. These relationships can be one-to-one or many-to-one. When obtaining each IMU data correction value, it is only necessary to input the motion influence deviation index into the mapping relationship table, and the motion attitude database can quickly locate and return each IMU data correction value corresponding to the motion influence deviation index. The larger the motion influence deviation index, the larger the IMU data correction value. Each IMU data correction value is respectively summed with the IMU data to obtain the corrected IMU data. MU data; IMU sensor parameters include sensitivity and sampling frequency; obtain the unit motion influence deviation index from the preset motion posture database, the unit motion influence deviation index is a quantitative indicator reflecting the degree of deviation caused by the change of a single motion quantity; perform a proportion approximation operation on the motion influence deviation index and the unit motion influence deviation index, and then round the proportion approximation operation result to obtain the adjustment value of each IMU sensor parameter, and sum the adjustment value of each IMU sensor parameter with the IMU sensor parameter to obtain the corrected IMU sensor parameter; match the motion influence deviation index with the GNSS speed data correction value corresponding to each motion influence deviation index preset in the motion posture database to obtain the GNSS speed data correction value, and use the GNSS speed data correction value to sum the GNSS speed data after the quantitative determination of the multipath effect to obtain the corrected GNSS speed data.

[0042] In this embodiment, dynamic corrections are made to IMU data and GNSS velocity data based on the results of the quantified determination of motion impact, effectively improving the accuracy and stability of the positioning and numerical values ​​of human motion posture data. The characteristics of the motion cycle are matched with the motion impact deviation index, and the parameters of the IMU sensor and GNSS velocity data are flexibly adjusted, thereby effectively reducing the impact of motion on data accuracy. This application can identify high-frequency and low-frequency movements in real time and make corresponding corrections based on the motion characteristics, thereby improving the adaptability and accuracy of human motion posture data in complex motion environments.

[0043] Furthermore, the step of determining whether to dynamically correct the window time of the IMU data of each motion cycle based on the motion characteristics of the IMU data of each motion cycle includes: obtaining the window time range corresponding to each motion feature from a preset motion posture database; determining whether the window time of the IMU data of each motion cycle is within the window time range corresponding to the motion feature; if so, the corresponding window time classification determination result is recorded as normal without additional processing; otherwise, the corresponding window time classification determination result is recorded as abnormal, and the window time of the IMU data of the motion cycle is adjusted to the corresponding window time range to obtain the processed IMU data.

[0044] In this embodiment, a convolutional neural network is used to process IMU data. Through the multi-level abstraction capabilities of the convolutional layer, the convolutional neural network can capture high-level features in the IMU data, thereby improving the accuracy and robustness of motion or posture recognition, and is particularly suitable for complex motion pattern recognition. When faced with noisy or irregular data, the convolutional neural network can still effectively identify key features and automatically learn the noise pattern during training, thereby enhancing the system's adaptability to irregular data. By determining whether to perform dynamic window time correction based on the motion characteristics of the IMU data in each motion cycle and automatically correcting abnormal data, the reliability of the IMU data can be improved, data deviations caused by window time errors can be avoided, and the overall accuracy of human motion posture data can be improved.

[0045] Furthermore, the step of integrating the IMU data after window time correction and the GNSS data after motion impact quantification to obtain human motion posture data includes: dynamically correcting the IMU data weight and the GNSS data weight according to the credibility parameters of the IMU data after window time classification and the GNSS data after motion impact quantification, and the IMU data weight and the GNSS data weight are used to adjust the contribution of the IMU data and the GNSS data to the state estimation during the unscented Kalman filter integration process to optimize the final state estimation accuracy and reliability; using the unscented Kalman filter to integrate the IMU data after window time classification and the GNSS data after motion impact quantification according to the corrected IMU data weight and the corrected GNSS data weight to obtain human motion posture data.

[0046] The Unscented Kalman Filter (UKF) is considered an improvement over the Extended Kalman Filter (EKF). The core concept of the UKF is to use an unscented transformation to approximate the posterior probability distribution of the state of a nonlinear system. Compared to the EKF, the UKF computes the posterior mean and covariance of the state in a Gaussian nonlinear system with third-order accuracy, which is higher than the EKF. Furthermore, the UKF does not require the calculation of the Jacobian matrix, making it applicable to strongly nonlinear systems.

[0047] like Figure 4 As shown in FIG, it is an algorithm flow chart of the unscented Kalman filter provided in an embodiment of the present application, wherein the sigma point of the unscented Kalman filter is obtained as follows: ;

[0048] For nonlinear systems, the state estimation of the unscented Kalman filter and its covariance matrix You can update it by following the method below: ; ; ; .

[0049] Among them, k is the current iteration process, k-1 is the previous iteration process, n is the variable latitude, and a is the adjustment parameter that controls the distribution of sigma points. is the covariance matrix, for In the i-th column of the matrix, X is the state estimation matrix, Z is the observation noise, and X* is the estimation matrix of the next state.

[0050] Among them, the step of dynamically correcting the IMU data weight and the GNSS data weight according to the credibility parameters of the IMU data after the window time classification judgment and the GNSS data after the motion impact quantification judgment includes: obtaining the critical fusion residual, the critical sensor covariance and the critical sensor update frequency from the preset motion posture database; performing a proportion approach calculation on the critical fusion residual, the critical sensor covariance and the sensor update frequency with the fusion residual, the sensor covariance and the critical sensor update frequency respectively, and then using the credibility parameter contribution to weight the proportion approach calculation results and couple them to obtain the credibility of the IMU data after the window time classification judgment and the GNSS data after the motion impact quantification judgment. The credibility index represents the quantitative data of the influence of the credibility parameters on the credibility of each data. The credibility parameters include the residual after fusion of the IMU data after the window time classification judgment and the GNSS data after the motion influence quantification judgment, the sensor covariance and the sensor update frequency; the credibility index of the IMU data after the window time classification judgment is compared with the credibility index of the GNSS data after the motion influence quantification judgment. If the credibility index of the IMU data after the window time classification judgment is greater than the credibility index of the GNSS data after the motion influence quantification judgment, the corresponding data credibility quantification judgment result is recorded as the first credibility assessment result, and the credibility index of the IMU data after the window time classification judgment is compared with the credibility index of the GNSS data after the motion influence quantification judgment. The sum of the credibility indexes of the GNSS data after quantitative judgment is marked as the total credibility index, the result of the approximation calculation of the ratio of the credibility index of the IMU data after the window time classification judgment to the total credibility index is recorded as the IMU data weight, and the result of the approximation calculation of the ratio of the credibility index of the GNSS data after the motion impact quantitative judgment to the total credibility index is recorded as the GNSS data weight; if the credibility index of the IMU data after the window time classification judgment is less than the credibility index of the GNSS data after the motion impact quantitative judgment, the corresponding data credibility quantitative judgment result is recorded as the second credibility assessment result, and the credibility index of the IMU data after the window time classification judgment is recorded as the credibility index of the GNSS data after the motion impact quantitative judgment. The sum of the credibility indices is marked as the total credibility index, the result of the approximation calculation of the ratio of the credibility index of the IMU data after the window time classification judgment to the total credibility index is marked as the IMU data weight, and the result of the approximation calculation of the ratio of the credibility index of the GNSS data after the motion impact quantification judgment to the total credibility index is marked as the GNSS data weight; if the credibility index of the IMU data after the window time classification judgment is equal to the credibility index of the GNSS data after the motion impact quantification judgment, the corresponding data credibility quantification judgment result is recorded as the third credibility assessment result, and the IMU data weight and GNSS data weight are adjusted to the preset weight values ​​(i.e., IMU data weight = GNSS data weight = 0.5).

[0051] The credibility index of IMU data after window time classification and GNSS data after motion impact quantification is obtained as follows: ;

[0052] In the formula, when a=1, a represents the IMU data after window time classification and judgment, It represents the credibility index of IMU data after window time classification judgment. When a=2, a represents the GNSS data after the quantitative judgment of motion impact. Indicates the credibility index of GNSS data after the quantitative determination of motion impact, Represents the residual contribution after fusion, represents the sensor covariance contribution, Indicates the contribution of sensor update frequency, It represents the fusion residual corresponding to the a-th judgment. It is the difference between the estimated value after Kalman filter fusion and the actual observation value, reflecting the error size in the fusion process. It can be obtained by calculating the difference between the data before fusion and the data after fusion. The larger the fusion residual, the smaller the credibility index. represents the residual after critical fusion, It represents the sensor covariance corresponding to the a-th judgment. The larger the sensor covariance, the smaller the credibility index. It represents the statistical characteristics of the sensor measurement error, which can be obtained through the sensor's technical specifications or historical observation data. represents the critical sensor covariance, It indicates the sensor update frequency corresponding to the a-th judgment. The greater the sensor update frequency, the greater the credibility index. It indicates the number of times the sensor updates data at each time monitoring point, which can be directly obtained from the sensor log record. Indicates the critical sensor update frequency.

[0053] 、 and The contributions of the fusion residual, sensor covariance, and sensor update frequency, which are preset in the motion posture database, represent the degree of influence of the fusion residual, sensor covariance, and sensor update frequency on the credibility index, respectively. These values ​​can be directly obtained from the motion posture database. These relationships are organized into a "mapping set," a lookup table or corresponding rules. When a credibility parameter value is actually monitored, the corresponding value can be searched in the corresponding mapping set to obtain the corresponding contribution. This contribution is a number between 0 and 1, representing the assessment of the importance of the current credibility parameter to the credibility index. For example, for the fusion residual, a mapping set exists between the fusion residual and the corresponding contribution. Entering the fusion residual value yields the corresponding contribution. For the sensor covariance, a mapping set exists between the sensor covariance and the corresponding contribution. Entering the sensor covariance value yields the corresponding contribution. For the sensor update frequency, a mapping set exists between the sensor update frequency and the corresponding contribution. Entering the sensor update frequency value yields the corresponding contribution. These mapping relationships can be many-to-one or one-to-one.

[0054] It should be noted that in the algorithm of this application, the post-fusion residual, sensor covariance, and sensor update frequency are interrelated. For example, a higher sensor update frequency can provide more refined dynamic information, which usually helps to reduce sensor covariance and improve system response, but may also increase noise. A larger sensor covariance indicates inaccurate measurement and larger error, which may lead to a larger post-fusion residual. A smaller sensor covariance indicates more accurate measurement and a smaller post-fusion residual.

[0055] In this embodiment, the data credibility index obtained through comprehensive analysis can accurately reflect the reliability of sensor data, helping the system to select the most reliable data for further processing in complex environments, thereby improving the accuracy of human motion posture data. At the same time, by dynamically adjusting the weight of the fused data, the data fusion process can be dynamically optimized, making the final result more accurate and reliable, and adaptable to different sensor performance and external environmental changes.

[0056] In summary, this embodiment quantitatively determines the multipath effect in GNSS signals and the motion-affecting parameters of each joint in the IMU sensor's motion cycle, and determines whether to dynamically correct the GNSS and IMU data, thereby reducing the impact of multipath and motion state on the data. Furthermore, the IMU data is dynamically corrected for window time based on the motion characteristics during the information extraction process, reducing the impact of window time differences on the efficiency of convolutional neural network feature extraction. Ultimately, by integrating the corrected IMU and GNSS data, high-precision human motion posture data acquisition is achieved.

[0057] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0058] The present invention is described with reference to flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0059] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0060] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1A step that specifies a function in one or more boxes.

[0061] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0062] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A human motion posture data acquisition system based on convolutional neural network, characterized by: include: Signal error correction module, motion impact correction module, data integration module and motion posture database; The signal error correction module is configured to quantitatively determine the impact of multipath effects based on signal error parameters during data acquisition by the GNSS sensor, obtain a quantitative determination result of the multipath effects, and determine whether to perform dynamic correction of the GNSS data based on the quantitative determination result of the multipath effects. The dynamic correction of the GNSS data refers to correcting the GNSS data based on the signal error parameters to reduce the impact of the multipath effects on the GNSS data. The GNSS data is GNSS position data and GNSS velocity data collected by the GNSS sensor. The motion influence correction module is used to perform a quantitative determination of the motion influence based on the motion influence parameters of each joint in each motion cycle during the data acquisition process of the IMU sensor, obtain a motion influence quantitative determination result, and determine whether to perform dynamic correction of the IMU data and dynamic correction of the GNSS speed data based on the motion influence quantitative determination result. The dynamic correction of the IMU data means correcting the IMU data according to the motion influence parameters to reduce the influence of the motion state on the IMU data. The dynamic correction of the GNSS speed data means correcting the GNSS speed data according to the motion influence parameters to reduce the influence of the motion state on the GNSS speed data. The data integration module is used to determine whether to perform dynamic window time correction on the IMU data of each motion cycle based on the motion characteristics of the IMU data of each motion cycle, and integrate the IMU data after the window time correction determination and the GNSS data after the motion impact quantification determination to obtain human body motion posture data. The dynamic window time correction means correcting the window time of the IMU data of each motion cycle based on the motion characteristics of the IMU data of each motion cycle to reduce the impact of the window time difference on the efficiency of the convolutional neural network feature extraction process.

2. The human motion posture data acquisition system based on a convolutional neural network as claimed in claim 1, characterized in that: The step of performing a quantitative determination of the multipath effect influence based on the signal error parameter in the data acquisition process of the GNSS sensor to obtain a quantitative determination result of the multipath effect influence comprises: Obtain signal error parameter reference data from a preset motion attitude database, specifically including: critical carrier phase residual, critical pseudorange multipath error, and critical signal-to-noise ratio; Performing a proportion approximation operation on each signal error parameter and the corresponding signal error parameter reference data, and then weighting and coupling the proportion approximation operation results using the signal error parameter contribution parameters to obtain a signal error index, wherein the signal error parameter contribution includes the carrier phase residual contribution, the pseudorange multipath error contribution, and the signal-to-noise ratio contribution. The signal error index represents quantitative data on the degree of influence of the signal error parameters on the accuracy of the GNSS data, wherein the signal error parameters include the carrier phase residual, the pseudorange multipath error, and the signal-to-noise ratio; Obtaining a first signal error threshold and a second signal error threshold from a preset motion posture database; Comparing the signal error index with the first signal error threshold and the second signal error threshold respectively; if the signal error index is less than or equal to the first signal error threshold, recording the corresponding multipath effect quantitative determination result as the first signal error result; If the signal error index is greater than the first signal error threshold and less than or equal to the second signal error threshold, the corresponding multipath effect impact quantization determination result is recorded as the second signal error result; If the signal error index is greater than the second signal error threshold, the corresponding multipath effect impact quantization determination result is recorded as the third signal error result.

3. The human motion posture data acquisition system based on convolutional neural network as claimed in claim 2, characterized in that: The step of determining whether to perform dynamic correction of GNSS data based on the quantitative determination result of the multipath effect includes: If the multipath effect affects the quantization determination result as the first signal error result, no additional processing is performed; If the quantified determination result of the multipath effect is the second signal error result, the difference between the second signal error threshold and the signal error index is recorded as the signal error deviation index, the signal error deviation index is matched with the GNSS data correction value corresponding to each signal error deviation index preset in the motion posture database to obtain the GNSS data correction value, and the GNSS data is corrected according to the GNSS data correction value to obtain the corrected GNSS data; If the multipath effect affects the quantitative determination result as the third signal error result, a GNSS data abnormality prompt is issued to determine whether signal filtering has been turned on. If so, the filtering noise parameter is adjusted according to the difference between the signal error index and the second signal error threshold. Otherwise, signal filtering is turned on. The filtering noise parameter includes process noise and observation noise.

4. The human body motion posture data acquisition system based on convolutional neural network as claimed in claim 3, characterized in that: The step of adjusting the filtering noise parameter according to the difference between the signal error index and the second signal error threshold comprises: Obtaining empirical coefficients from a preset motion posture database; The difference between the signal error index and the second signal error threshold is marked as the signal error deviation amplitude, and matched with the process noise multiples corresponding to each signal error deviation amplitude preset in the motion posture database to obtain the process noise multiple, and the process noise is adjusted according to the process noise multiple; The observation noise is adjusted according to the current observation noise, the empirical coefficient and the signal error deviation amplitude.

5. The human motion posture data acquisition system based on convolutional neural network according to claim 1, characterized in that: The step of performing quantitative determination of motion influence based on the motion influence parameters of each joint in each motion cycle during data acquisition by the IMU sensor to obtain the quantitative determination result of motion influence comprises: Obtain motion influencing parameter reference data from a preset motion posture database, specifically including: critical acceleration amplitude, critical angular velocity amplitude, and critical average acceleration; The motion influence parameters of each joint in each motion cycle are respectively subjected to proportion approximation calculation with the corresponding motion influence parameter reference data, and then the proportion approximation calculation results are weighted and coupled using the motion influence parameter contribution, and then the coupling processing results are averaged to obtain the motion influence index of each motion cycle, wherein the motion influence parameter contribution includes the acceleration amplitude contribution, the angular velocity amplitude contribution and the average acceleration contribution, and the motion influence index represents the quantitative data of the degree of influence of the motion influence parameters on the accuracy of the IMU data, and the motion influence parameters include the acceleration amplitude, the angular velocity amplitude and the average acceleration; Obtaining a motion impact threshold from a preset motion posture database, and comparing the motion impact index of each motion cycle with the motion impact threshold; If the motion impact index of a certain motion cycle does not exceed the motion impact threshold, the corresponding motion impact quantification determination result is marked as the first motion impact result; If the motion impact index of a certain motion cycle exceeds the motion impact threshold, the corresponding motion impact quantification determination result is marked as the second motion impact result.

6. The human motion posture data acquisition system based on convolutional neural network according to claim 5, characterized in that: The step of determining whether to perform dynamic correction of IMU data and dynamic correction of GNSS velocity data based on the quantified result of motion impact includes: If the quantified result of the motion influence of a certain motion cycle is the first motion influence result, the motion feature of the motion cycle is marked as a low-frequency motion and no additional processing is performed; If the quantitative determination result of the motion impact of a certain motion cycle is the second motion impact result, the motion feature of the motion cycle is marked as a high-frequency motion, and the difference between the motion impact index of the motion cycle and the motion impact threshold is marked as the motion impact deviation index. The IMU data is dynamically corrected according to the motion impact deviation index, and the GNSS speed data is dynamically corrected.

7. The human motion posture data acquisition system based on convolutional neural network according to claim 6, characterized in that: The steps of dynamically correcting the IMU data according to the motion influence deviation index and dynamically correcting the GNSS velocity data include: Matching the motion influence deviation index with the IMU data correction value corresponding to each motion influence deviation index preset in the motion posture database to obtain the IMU data correction value, and correcting the IMU data according to the IMU data correction value to obtain the corrected IMU data; Obtaining a unit motion impact deviation index from a preset motion posture database, wherein the unit motion impact deviation index is a quantitative indicator reflecting the degree of deviation caused by a change in a single motion amount; Performing a ratio approximation operation on the motion influence deviation index and the unit motion influence deviation index, and then performing integer processing on the ratio approximation operation result to obtain an IMU sensor parameter adjustment value, and adjusting the IMU sensor parameters according to the IMU sensor parameter adjustment value; The motion influence deviation index is matched with the GNSS speed data correction value corresponding to each motion influence deviation index preset in the motion attitude database to obtain the GNSS speed data correction value. The GNSS speed data correction value is used to correct the GNSS speed data after the quantitative determination of the multipath effect to obtain the corrected GNSS speed data.

8. The human motion posture data acquisition system based on convolutional neural network according to claim 6, characterized in that: The step of determining whether to perform window time dynamic correction on the IMU data of each motion cycle according to the motion characteristics of the IMU data of each motion cycle includes: Obtain the window time range corresponding to each action feature from the preset motion posture database; Determine whether the window time of the IMU data of each motion cycle is within the window time range corresponding to the motion feature. If so, the corresponding window time classification judgment result is recorded as normal and no additional processing is performed. Otherwise, the corresponding window time classification judgment result is recorded as abnormal, and the window time of the IMU data of the motion cycle is adjusted to the corresponding window time range to obtain the processed IMU data.

9. The human motion posture data acquisition system based on convolutional neural network according to claim 1, characterized in that: The step of integrating the IMU data after window time correction and the GNSS data after motion impact quantification to obtain human motion posture data includes: Dynamically modify the IMU data weight and GNSS data weight based on the credibility parameters of IMU data after window time classification and GNSS data after motion impact quantification. The unscented Kalman filter is used to integrate the IMU data after window time classification and the GNSS data after motion impact quantification according to the corrected IMU data weights and the corrected GNSS data weights to obtain the human body motion posture data.

10. The human motion posture data acquisition system based on convolutional neural network according to claim 9, characterized in that: The step of dynamically correcting the IMU data weight and the GNSS data weight according to the credibility parameters of the IMU data after the window time classification determination and the GNSS data after the motion impact quantification determination comprises: Obtain critical fusion residual, critical sensor covariance and critical sensor update frequency from a preset motion posture database; Performing a proportion approximation operation on the critical fusion residual, critical sensor covariance, and sensor update frequency with the fusion residual, sensor covariance, and critical sensor update frequency, respectively. Then, using the credibility parameter contribution, weighting and coupling processing are performed on the proportion approximation operation results to obtain the credibility index of the IMU data after window time classification and the GNSS data after motion effect quantification. The credibility index represents the quantitative data of the degree of influence of the credibility parameters on the credibility of each data. The credibility parameters include the fusion residual, sensor covariance, and sensor update frequency of the IMU data after window time classification and the GNSS data after motion effect quantification; Compare the credibility index of the IMU data after the window time classification judgment with the credibility index of the GNSS data after the motion impact quantification judgment. If the credibility index of the IMU data after the window time classification judgment is greater than the credibility index of the GNSS data after the motion impact quantification judgment, record the corresponding data credibility quantification judgment result as the first credibility assessment result, mark the sum of the credibility index of the IMU data after the window time classification judgment and the credibility index of the GNSS data after the motion impact quantification judgment as the total credibility index, record the result of the approximation of the proportion of the credibility index of the IMU data after the window time classification judgment to the total credibility index as the IMU data weight, and record the result of the approximation of the proportion of the credibility index of the GNSS data after the motion impact quantification judgment to the total credibility index as the GNSS data weight; If the credibility index of the IMU data after the window time classification determination is less than the credibility index of the GNSS data after the motion impact quantification determination, the corresponding data credibility quantification determination result is recorded as the second credibility assessment result, the sum of the credibility index of the IMU data after the window time classification determination and the credibility index of the GNSS data after the motion impact quantification determination is marked as the total credibility index, the result of the approximation calculation of the proportion of the credibility index of the IMU data after the window time classification determination and the total credibility index is marked as the IMU data weight, and the result of the approximation calculation of the proportion of the credibility index of the GNSS data after the motion impact quantification determination and the total credibility index is marked as the GNSS data weight; If the credibility index of the IMU data after the window time classification judgment is equal to the credibility index of the GNSS data after the motion impact quantification judgment, the corresponding data credibility quantification judgment result is recorded as the third credibility assessment result, and the IMU data weight and GNSS data weight are adjusted to the preset weight values.

Images