Research On Human Gait Pattern Recognition Based On Human-Machine Coordination

Recently,due to ageing societal concerns,new scenarios for providing assistance to elderly people for carrying out daily living activities have also started to receive attention,and there is growing urgency for such assistive technologies to help elderly people remain independent.In such situations,the body-assisted exoskeleton robot can provide corresponding assistance,improve the mobility of the elderly,support the elderly to complete basic daily activities,and alleviate the aging of the population?The accurate gait pattern recognition is the premise and basis for the determination of the motion intent and the establishment of control strategies of the exoskeleton robot.In this study,we developed a wearable,multi-sensor-based human gait information acquisition system.The signals from the left and right legs of the four inertial sensors(IMU)are collected via the CAN-bus and displayed and stored by the host computer.Then the signal segment of the acquired acceleration and angular acceleration of each joint of the lower extremity is windowed to extract the time-domain features and frequency-domain features from a single time window.In order to improve the robustness and accuracy of the algorithm,a 6-person gait data set was constructed.Finally,using different classifiers for classification and recognition,based on the three algorithms of SVM,kNN,and DTs,A set of gait feature extraction and machine learning algorithms based on kNN and DTW are proposed.The experimental results show that the feature extraction method combining time domain and frequency domain can effectively distinguish 4 types of standing,walking horizontally,going up stairs and down stairs.Different gait patterns,when the window size is 300ms,the average recognition rate of kNN&DTW classifier reaches 98.23%.