Gait Analysis And Research Based On Multi-MEMS Inertial Sensors

As the most common form of exercise,walking involves the knowledge of kinesiology,dynamics and physiology.Because of the differences in height,weight and control of the nervous system,each person's gait is unique,so gait analysis can be used in rehabilitation assessment,identity recognition,sports and other fields.The gait analysis and its application is one of the most popular research directions.This thesis proposes a MEMS gait analysis method based on inertial sensors.A self-made inertial motion capture module was used to perform real-time motion capture and attitude calculations.The motion data was uploaded to the host computer using Bluetooth communication for multi-node data fusion and motion visualization.Gait parameter calculation and feature extraction were performed.The correlation between gait parameters and the relationship between gait information and gait recognition was studied.The main contents are given as follows:1)An inertial sensor-based motion capture module was designed and implemented for real-time acquisition of gait information.The module uploaded the acquired data to the host computer for storage.The hardware part of this module includes microprocessor module,MPU6050 module,HMC5883 L module,wireless communication module,power supply module and peripheral interface module.The software part mainly implemented the low-level circuit driver,accelerometer and magnetometer calibration,metadata reading and filtering,quaternion attitude fusion and wireless communication.The motion capture module can acquire the motion attitude angle,triaxial angular velocity,triaxial acceleration,and triaxial magnetometer signals,providing the basis for the following study.2)A three-dimensional human body hierarchy skeleton model was created.The motion data of each node was correlated with the three-dimensional human body model and was stored as a BVH(Biovision Hierarchy)file format.Using C# and OpenGL library,a motion visualization software that supports the BVH file format was developed.3)The sensor data was merged,the starting point of each phase in the gait cycle was determined and the gait parameters(gait time-space parameters and symmetry parameters)were calculated.To reduce the gait misjudgment caused by the random noise of sensor data and the gait error detection under different motion patterns,an adaptive threshold gait detection and misjudgment correction algorithm was proposed to classify the stationary and dynamic phase in the gait cycle.A zero-velocity update algorithm was used in the stationary state of the gait to correct the speed of the motion and improve the accuracy of the displacement estimation.Phase-rectified Signal Averaging(PRSA)was used to extract gait features from the collected gait data.4)Gait analysis was performed on thirty-six volunteers Two hundred and forty-eight gait samples were obtained.Pearson correlation coefficient was used to analyze the correlation of gait parameters of volunteers,and the correlation coefficient between various gait parameters was obtained.Two sets of characteristic parameters were extracted: 1.According to the correlation,the gait parameters were selected to form a 10-dimensional gait information vector;2.The gait acceleration,angular velocity,and swing amplitude were extracted using the PRSA algorithm.The gait features and gait parameters form a 20-dimensional gait feature vector.Using a support vector machine to learn gait information,a mapping relationship between gait information and age groups was established to identify young people and middle-aged people.173 samples were selected as training samples and the remaining 75 samples were used as the test sets.The accuracy of the model using the 10-dimensional gait information vector as the feature parameter was 86.70%,and the accuracy rate of the model using the 20-dimensional gait information vector was 90.55%.The experiment proves that the gait parameters and features acquired by the gait analysis system can meet the needs of gait recognition.The gait characteristics obtained by the PRSA feature extraction algorithm and the gait parameters are used as identification factors to identify the identity compared to single step state parameters.The accuracy rate of identification was increased by 3.85%.