Hierarchical finite-state machine gait recognition method based on multi-source data fusion and assistance control system

Abstract

The invention discloses a hierarchical finite-state machine gait recognition method based on multi-source data fusion and a power-assisted control system, and particularly relates to the technical field of wearable power-assisted equipment control. According to the method, refined recognition of gait phases is achieved by constructing a three-level state machine architecture of a motion mode layer, a gait phase layer and a micro-phase layer; plantar pressure sensing data and thigh inertial measurement unit data are creatively fused, and the problems of noise interference and detection blind areas of a single sensor are solved by using the time-space complementary characteristic; sliding window filtering is adopted to preprocess a pressure signal, and an IMU dip angle compensation mechanism is designed to improve the robustness of the system. The system is used for executing the method, the torque output precision and real-time performance of the wearable power assisting device are remarkably optimized, the recognition delay is smaller than 35 ms, the misjudgment rate is smaller than 5%, the system is suitable for real-time control over power assisting devices such as lower limb exoskeletons and intelligent prostheses, and the recognition precision and robustness are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wearable assistive device control, more particularly, the present application relates to a hierarchical finite state machine gait recognition method based on multi-source data fusion and an assistive control system. BACKGROUND

[0002] Gait recognition is a key technology for wearable assistive devices to realize human-machine collaboration. The existing technical solutions mainly have the following defects: 1. Single sensor solution has insufficient reliability. For example, a solution using only an IMU (Inertial Measurement Unit) is susceptible to limb swing noise interference, has a high misjudgment rate, and a single pressure sensor has a data blind area during the foot-off ground stage. Secondly, the traditional state machine divides the gait phase too roughly and does not subdivide the support period, resulting in confusion of torque output strategies in different scenarios such as walking on flat ground and going up and down stairs, which causes user gait disorder. Furthermore, the real-time performance of deep learning recognition algorithms based on multi-source data fusion on embedded platforms is poor, making it difficult to meet the millisecond-level response requirements of gait control.

[0003] Therefore, a hierarchical finite state machine gait recognition method based on multi-source data fusion and an assistive control system are proposed as further improvements to provide a high-precision, high-real-time, and high-robustness gait recognition method and system. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a hierarchical finite state machine gait recognition method based on multi-source data fusion and an assistive control system to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions: a hierarchical finite state machine gait recognition method based on multi-source data fusion, comprising the following steps: S1: collecting user plantar pressure sensor data and thigh inertial measurement unit data through sensors; S2: synchronizing and preprocessing the data collected in S1: S3: top-level recognition of the data processed in S2: recognizing the user's state according to the thigh inertial measurement unit data to determine which macro motion mode it is in; S4: top-level recognition: according to the judgment of S3, enter the corresponding macro motion mode; then enter S5; S5: middle-level recognition: recognize the user's gait phase according to the plantar pressure sensor data; the gait phase includes at least a support period and a swing period; if in the support period, enter S6; if in the swing period, enter S7; S6: Perform bottom layer identification: divide the foot bottom into three areas, according to the area where the pressure center is located, drive the power-assisted device to execute the corresponding power-assisted strategy; then return to S1; S7: The power-assisted device stops assisting.

[0006] Further, in S2, the sensor data synchronization includes the following operations: S21: Synchronize the foot bottom pressure voltage values of multiple channels; S22: Read the thigh inertial measurement unit data containing three-axis acceleration (ax, ay, az) and three-axis angular velocity (gx, gy, gz) of each frame.

[0007] Further, in S2, data preprocessing: for the collected foot bottom pressure data, use the sliding window mean filtering method for preprocessing.

[0008] Further, in S3, the macro motion mode is divided into If σ 2 <0.5 rad 2 / s 2 , it is determined that the current is a static mode; If σ 2 ∈ [1.2, 3.0] rad 2 / s 2 , it is determined that the current is a flat walking mode; If σ 2 >4.0 rad 2 / s 2 , it is determined that the current is a running mode.

[0009] Further, in S4, the top-level state machine outputs the mode identification of the macro motion mode to the middle-level state machine and the bottom-level state machine.

[0010] Further, in S5, the middle-level state machine receives the mode identification output by the top-level state machine and the filtered heel pressure value in S2, and performs support period and swing period determination, with the following judgment conditions: If the state of heel pressure value > 5N lasts more than 20ms, it is determined to enter the support period; In the support period, calculate the forward movement speed of the pressure center, if the forward movement speed of the pressure center > 0.2 m / s, it is determined to enter the swing period; If no valid heel pressure signal is received for 200ms, enable IMU compensation judgment, calculate the thigh inclination angle θ = arctan(ax / az), if θ < 10° at this time, it is determined to be in the support period; otherwise, it is determined to be in the swing period.

[0011] Furthermore, in S6, the length of the sole is defined as L, and the sole is divided into three regions: the heel region [0, L / 3], the midfoot region [L / 3, 2L / 3], and the toe region [2L / 3, L]. If the center of pressure ∈ [0, L / 3], the heel touches the ground, and during this stage, the power assist motor outputs peak torque mode; If the center of pressure ∈ [L / 3, 2L / 3], then the entire foot is on the ground. During this stage, the output torque of the assist motor decreases linearly according to the position of the center of pressure.

[0012] If the center of pressure ∈ [2L / 3, L], then the toes touch the ground; during this stage, when the forefoot pressure value is >3N, the output torque of the power assist motor decreases linearly according to the pressure value; when the forefoot pressure value is <3N, the power assist motor output is turned off, and the output torque of the power assist motor drops to zero.

[0013] An assistive control system includes a hierarchical finite state machine gait recognition method based on multi-source data fusion, and further includes: Pressure sensing module, which is used to collect plantar pressure data; The IMU sensing module is used to collect thigh inertial data; The main control module, whose signal input terminals are communicatively connected to the signal output terminals of the pressure sensing module and the IMU sensing module, is used to execute the hierarchical finite state machine gait recognition method and output gait state commands; and, The communication module has its signal input terminal connected to the signal output terminal of the main control module, and is used to transmit gait status commands to the actuator of the assist device.

[0014] Furthermore, the main control module uses an STM32F407 chip, whose built-in timer is configured to trigger the ADC and serial port for synchronous sampling between the pressure sensing module and the IMU sensing module.

[0015] Furthermore, the pressure sensing module is a multi-point thin-film pressure sensor array arranged on the sole of the foot; The sole includes: the insole heel area, the insole midfoot area, the insole left forefoot area, and the insole right forefoot area; the insole heel area corresponds to the heel area, the insole midfoot area corresponds to the midfoot area, and the insole left forefoot area and the insole right forefoot area correspond to the toe area. The pressure sensing module collects foot pressure sensing data in the corresponding area of ​​the user's foot. The IMU sensing module is installed on the middle of the outer side of the user's thigh to measure the user's thigh movement posture and collect thigh inertial measurement unit data.

[0016] The technical effects and advantages of this invention are as follows: 1. Regarding recognition accuracy and robustness: This invention innovatively integrates user foot pressure sensor data and thigh inertial measurement unit (IMU) data, utilizing the spatiotemporal complementary characteristics of the foot pressure sensor's high accuracy during the support phase and the IMU's blind-zone-free operation during the swing phase. This is expected to significantly overcome the problems of single-sensor solutions being susceptible to noise interference and having detection blind zones. Based on reasonable inference, the average misclassification rate of the method under various gait conditions is expected to be reduced to below 5%, a significant improvement compared to the traditional single-IMU solution (misclassification rate 15%-20%).

[0017] 2. Regarding the precision of assist timing: Utilizing a three-layered state machine architecture of 'movement mode - gait phase - micro-phase', it achieves refined identification of four micro-phases during the support phase: heel strike, full foot strike, toe strike, and toe lift. This is expected to provide the assist motor with more advanced and precise control timing. This design is anticipated to improve the alignment between the assist torque output and the user's muscle power burst point, thereby more effectively reducing the user's metabolic consumption and improving the naturalness of human-machine coordination.

[0018] 3. In terms of system real-time performance: By adopting threshold-based temporal feature extraction (such as variance calculation and threshold comparison) and state machine logic judgment to replace the computationally complex deep learning model, all algorithms are optimized for embedded MCUs such as STM32F407. It is expected that the delay of the entire recognition process can be controlled within 35ms, which fully meets the timeliness requirements of real-time assistance in the gait cycle. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method steps of the present invention.

[0020] Figure 2 This is a three-level state machine transition logic diagram according to an embodiment of the present invention.

[0021] Figure 3 This is a system block diagram of the present invention.

[0022] The attached diagram is labeled as follows: 1. Pressure sensing module; 2. IMU sensing module; 3. Main control module; 4. Communication module; 5. Assistive device actuator. Detailed Implementation

[0023] 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. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0024] As attached Figure 1The method for gait recognition based on hierarchical finite state machine fusion, as shown, includes the following steps: S1: Collects user's foot pressure sensing data and thigh inertial measurement unit data through sensors; Among them, the sensor that collects user plantar pressure sensing data is used to identify the user's macroscopic movement pattern, and this sensor can be selected as a plantar pressure sensor. Among them, the sensor that collects data from the user's thigh inertial measurement unit is used to identify the user's gait phase, and this sensor can be selected as an inertial measurement unit (IMU). S2: Synchronize and preprocess the data collected by S1: For example, sensor data synchronization includes the following operations: S21: Simultaneously acquire plantar pressure voltage values ​​from multiple channels; S22: Read the thigh inertial measurement unit data for each frame, which includes triaxial acceleration (ax, ay, az) and triaxial angular velocity (gx, gy, gz); For example: Data preprocessing: For the collected plantar pressure data, a sliding window mean filtering method is used for preprocessing. The window width is set to 5 sampling points (corresponding to a 50ms time window). The filtered value (P_filtered(t)) at the current time t and for each pressure point is calculated using the formula: P_filtered(t) = (P(t) + P(t-1) + P(t-2) + P(t-3) + P(t-4)) / 5 S3: Perform top-level identification on the data processed by S2: Identify the user's state based on the thigh inertial measurement unit data and determine the macroscopic motion mode in which the user is in order to select the corresponding assist strategy based on the identified macroscopic motion mode; wherein, the macroscopic motion mode is divided by calculating the angular velocity variance of the thigh inertial measurement unit data, for example, the macroscopic motion modes include: stationary state, flat ground walking state and running state. More specifically: The macroscopic motion pattern is divided by calculating the variance of the angular velocity of the thigh inertial measurement unit data in S22; for example, calculating the variance σ of the angular velocity gy (Y-axis angular velocity, corresponding to sagittal plane oscillation) of the thigh inertial measurement unit data within the most recent 100ms (N=10 sampling points). 2 ; If σ 2 <0.5 rad 2 / s 2 If so, the current mode is determined to be static. If σ 2 ∈ [1.2, 3.0] rad 2 / s 2 If so, the current mode is determined to be flat ground walking mode; If σ 2 >4.0 rad 2 / s 2 If so, the current mode is determined to be running.

[0025] S4: Perform top-level recognition: Based on the judgment in S3, enter the corresponding macroscopic motion mode; wherein, the top-level state machine outputs the mode identifier of the macroscopic motion mode to the middle-level state machine and the bottom-level state machine; then proceed to S5; S5: Perform mid-level identification: Identify the user's gait phase based on plantar pressure sensor data; the gait phase includes at least the support phase and the swing phase; if it is in the support phase, proceed to S6; if it is in the swing phase, proceed to S7; Among them, gait phase is determined by the threshold and duration of plantar pressure value; if it is during the support phase, it is further subdivided into multiple micro-phases based on the change trajectory of the user's plantar pressure sensor data. More specifically, after receiving the mode identifier output by the top-level state machine and the filtered heel pressure value in S2, the middle-level state machine determines the support phase and the swing phase, based on the following conditions: If the heel pressure value is greater than 5N for more than 20ms, it is determined that the support phase has begun. During the support phase, the forward movement velocity of the pressure center is calculated. If the forward movement velocity of the pressure center is >0.2 m / s, it is determined that the oscillation phase has begun. If no valid heel pressure signal is received for 200ms, indicating that the plantar pressure sensor data acquisition has failed, the gait phase is compensated and determined using the IMU tilt angle data: the thigh tilt angle θ = arctan(ax / az) is calculated. If θ < 10°, it is determined to be the support phase; otherwise, it is determined to be the swing phase. S6: Perform bottom-level identification: Divide the sole of the foot into three regions, and identify the micro-phase based on the region where the pressure center is located. The result will then drive the power assist device to execute the corresponding power assist strategy; then return to S1. Among them, the micro-phase is determined by the position of the trajectory of the center of pressure on the sole of the foot within the sole zone. Multiple micro-phases include: heel-to-ground micro-phase, full-foot-to-ground micro-phase, toe-to-ground micro-phase, and toe-to-ground-leaning micro-phase. For example: Define the length of the foot as L, and divide the foot into three regions: the heel region [0, L / 3], the midfoot region [L / 3, 2L / 3], and the toe region [2L / 3, L]. If the center of pressure ∈ [0, L / 3], then the heel touches the ground, which is the micro-phase of the heel touch. During this phase, the power assist motor outputs peak torque mode. If the pressure center ∈ [L / 3, 2L / 3], then the foot is fully on the ground and is in the full-foot-on-the-ground micro-phase. During this phase, the output torque of the assist motor decreases linearly according to the position of the pressure center.

[0026] If the center of pressure ∈ [2L / 3, L], then the toes touch the ground; during this phase, when the forefoot pressure value is >3N, it is in the micro-phase of toe contact with the ground, and the output torque of the power assist motor decreases linearly according to the pressure value; when the forefoot pressure value is <3N, it is in the micro-phase of toe off the ground, the power assist motor output is turned off, and the output torque of the power assist motor drops to zero; S7: The power assist device stops providing power.

[0027] Among them, S4, S5 and S6, top-level recognition, middle-level recognition and bottom-level recognition are the recognition processes of the hierarchical finite state machine. They are all executed periodically in the timer interrupt service routine or the main loop, with a period of 10ms (100Hz).

[0028] As attached Figure 3 As shown, an assistive control system includes a hierarchical finite state machine gait recognition method based on multi-source data fusion, and further includes: Pressure sensing module 1, which is used to collect plantar pressure data; IMU sensing module 2, which is used to collect thigh inertial data; Main control module 3, whose signal input terminals are communicatively connected to the signal output terminals of pressure sensing module 1 and IMU sensing module 2 respectively, is used to execute the hierarchical finite state machine gait recognition method and output gait state commands; and, Communication module 4, whose signal input terminal is connected to the signal output terminal of main control module 3, is used to transmit gait status commands to the actuator 5 of the assist device; For example: The core hardware of the system uses STMicroelectronics' STM32F407ZGT6 microcontroller as the main control module 3, with a working frequency of 168MHz; The pressure sensing module 1 uses a thin-film pressure sensor, which is distributed and attached to four areas: the heel of the insole, the middle of the insole, the left forefoot of the insole, and the right forefoot of the insole. After being conditioned by an operational amplifier circuit, the four pressure sensors are connected to the ADC1_IN1 to ADC1_IN4 pins of the main control module 3 respectively. IMU sensor module 2 uses the JY61 model inertial measurement unit from Witt Intelligent Technology Co., Ltd., which is installed and fixed on the middle of the outer side of the user's right thigh to measure the thigh's movement posture; IMU sensor module 2 is connected to the USART2_RX / USART2_TX pins of the main control module 3 through a serial communication interface, with the communication baud rate set to 115200bps; The motor control pins (such as TIM1_CH1, TIM1_CH2) of the main control module 3 are connected to the actuator 5 of the assist device to output assist control signals; wherein, the actuator 5 of the assist device adopts the motor driver of the lower limb exoskeleton.

[0029] The motor control pins (such as TIM1_CH1, TIM1_CH2) of the main control module 3 are connected to the communication module 4 of the lower limb exoskeleton, i.e. the motor driver. The communication module 4 is used to output assist control signals. Communication module 4 uses Copley Controls' ACJ-055-18 motor driver, which drives Maxon brushless motor EC-90 through the CANopen communication protocol to achieve position and torque control. The communication baud rate is set to 1Mbps. Maxon brushless motor EC-90 is the actuator 5 of the power assist device.

[0030] The assistive device actuator 5 provides assistive control output: the system combines the mode with micro-phase to form a complete gait state code; it utilizes the torque-phase curve mapping table of different movement modes pre-stored inside the assistive device actuator 5. Based on the Gait_State_Code identified in real time by the main control module 3, this table is queried, and when it outputs the corresponding signal to the communication module 4, it controls the assistive device actuator 5 of the exoskeleton to output precise assistive torque.

[0031] In a preferred embodiment, as shown in the appendix Figure 3 As shown, the main control module 3 uses an STM32F407 chip, whose built-in timer is configured to trigger the ADC and serial port for synchronous sampling of the pressure sensing module 1 and the IMU sensing module 2.

[0032] Among them, for sensor data synchronization, the general-purpose timer TIM3 of the main control module 3 is configured to operate at a frequency of 10kHz, and the following operation is triggered in its update interrupt service function: 1. Start the rule group conversion of ADC1 and simultaneously collect the plantar pressure voltage values ​​(V_p1 to V_p4) of the four channels. 2. Send a data request command to the IMU sensing module 2 via USART2 to read a frame of thigh inertial measurement unit data containing triaxial acceleration (ax, ay, az) and triaxial angular velocity (gx, gy, gz).

[0033] For the collected plantar pressure data, a sliding window mean filtering method was used for preprocessing. The window width was set to 5 sampling points (corresponding to a 50ms time window). The filtered value (P_filtered(t)) at the current time t and for each pressure point was calculated using the formula: P_filtered(t) = (P(t) + P(t-1) + P(t-2) + P(t-3) + P(t-4)) / 5 In a preferred embodiment, as shown in the appendix Figure 3 As shown, pressure sensing module 1 is a multi-point thin-film pressure sensor array arranged on the sole of the foot; The sole of the foot includes: the heel area of ​​the insole, the midfoot area of ​​the insole, the left forefoot area of ​​the insole, and the right forefoot area of ​​the insole; the heel area of ​​the insole corresponds to the heel area, the midfoot area of ​​the insole corresponds to the midfoot area, and the left forefoot area of ​​the insole and the right forefoot area of ​​the insole correspond to the toe area. Pressure sensing module 1 collects foot pressure sensing data in the corresponding area of ​​the user's foot; IMU sensor module 2 is installed on the middle of the outer side of the user's thigh to measure the user's thigh movement posture and collect thigh inertial measurement unit data.

[0034] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A hierarchical finite state machine gait recognition method based on multi-source data fusion, characterized in that: Includes the following steps: S1: Collects user's foot pressure sensing data and thigh inertial measurement unit data through sensors; S2: Synchronize and preprocess the data collected by S1: S3: Perform top-level identification on the data processed by S2: Identify the user's state based on the data from the thigh inertial measurement unit and determine the macroscopic motion mode in which the user is. S4: Perform top-level recognition: Based on the judgment in S3, enter the corresponding macroscopic motion mode; then proceed to S5; S5: Perform mid-level identification: Identify the user's gait phase based on plantar pressure sensor data; the gait phase includes at least the support phase and the swing phase; if it is in the support phase, proceed to S6; if it is in the swing phase, proceed to S7; S6: Perform bottom-level recognition: Divide the sole of the foot into three regions, and drive the assist device to execute the corresponding assist strategy according to the region where the pressure center is located; then return to S1; S7: The power assist device stops providing power.

2. The hierarchical finite state machine gait recognition method based on multi-source data fusion according to claim 1, characterized in that: In step S2, sensor data synchronization includes the following operations: S21: Simultaneously acquire plantar pressure voltage values ​​from multiple channels; S22: Read thigh inertial measurement unit data for each frame, which includes triaxial acceleration (ax, ay, az) and triaxial angular velocity (gx, gy, gz).

3. The hierarchical finite state machine gait recognition method based on multi-source data fusion according to claim 2, characterized in that: In step S2, data preprocessing is performed: the collected plantar pressure data is preprocessed using a sliding window mean filtering method.

4. The hierarchical finite state machine gait recognition method based on multi-source data fusion according to claim 3, characterized in that: In S3, the macroscopic motion mode is divided by calculating the angular velocity variance of the thigh inertial measurement unit data in S22; If σ 2 < 0.5 rad 2 / s 2 If so, the current mode is determined to be static. If σ 2 ∈ [1.2, 3.0] rad 2 / s 2 If so, the current mode is determined to be flat ground walking mode; If σ 2 > 4.0 rad 2 / s 2 If so, the current mode is determined to be running.

5. The hierarchical finite state machine gait recognition method based on multi-source data fusion according to claim 4, characterized in that: In S4, the top-level state machine outputs the mode identifier of the macroscopic motion mode to the middle-level state machine and the bottom-level state machine.

6. The hierarchical finite state machine gait recognition method based on multi-source data fusion according to claim 5, characterized in that: In step S5, the middle-level state machine receives the mode identifier output by the top-level state machine and the filtered heel pressure value from step S2, and then determines the support phase and the swing phase based on the following conditions: If the heel pressure value is >5N for more than 20ms, it is determined that the support phase has begun; During the support phase, the forward movement velocity of the pressure center is calculated. If the forward movement velocity of the pressure center is > 0.2 m / s, it is determined that the oscillation phase has begun. If no effective heel pressure signal is received for 200ms, IMU compensation decision is enabled, and the thigh tilt angle θ = arctan(ax / az) is calculated. If θ < 10°, it is determined to be the support phase; otherwise, it is determined to be the swing phase.

7. The hierarchical finite state machine gait recognition method based on multi-source data fusion according to claim 6, characterized in that: In S6, the length of the sole is defined as L, and the sole is divided into three regions: the heel region [0, L / 3], the midfoot region [L / 3, 2L / 3], and the toe region [2L / 3, L]. If the center of pressure ∈ [0, L / 3], the heel touches the ground, and during this stage, the power assist motor outputs peak torque mode; If the center of pressure ∈ [L / 3, 2L / 3], then the entire foot is on the ground. During this stage, the output torque of the assist motor decreases linearly according to the position of the center of pressure. If the center of pressure ∈ [2L / 3, L], then the toes touch the ground; during this stage, when the forefoot pressure value > 3N, the output torque of the power assist motor decreases linearly according to the pressure value; when the forefoot pressure value < 3N, the power assist motor output is turned off, and the output torque of the power assist motor drops to zero.

8. A power assist control system, comprising the hierarchical finite state machine gait recognition method based on multi-source data fusion as described in claim 7, characterized in that: include: Pressure sensing module (1), which is used to collect plantar pressure data; IMU sensing module (2), which is used to collect thigh inertial data; The main control module (3) has its signal input terminals connected to the signal output terminals of the pressure sensing module (1) and the IMU sensing module (2), respectively. The main control module (3) is used to execute the hierarchical finite state machine gait recognition method and output gait state commands; and... The communication module (4) is connected to the signal output of the main control module (3) for transmitting gait status commands to the actuator (5) of the assist device.

9. The power assist control system according to claim 8, characterized in that: The main control module (3) uses an STM32F407 chip, whose built-in timer is configured to trigger the ADC and serial port for synchronous sampling of the pressure sensing module (1) and the IMU sensing module (2).

10. A power assist control system according to claim 8, characterized in that: The pressure sensing module (1) is a multi-point thin-film pressure sensor array arranged on the sole of the foot; The sole includes: the insole heel area, the insole midfoot area, the insole left forefoot area, and the insole right forefoot area; the insole heel area corresponds to the heel area, the insole midfoot area corresponds to the midfoot area, and the insole left forefoot area and the insole right forefoot area correspond to the toe area. The pressure sensing module (1) collects foot pressure sensing data in the corresponding area of ​​the user; The IMU sensing module (2) is installed on the middle of the outer side of the user's thigh to measure the user's thigh movement posture and collect thigh inertial measurement unit data.

Images