Recognition And Prediction Of Human Lower Limb Movements Based On Wearable Sensors

With the rapid development of science and technology,lower extremity exoskeleton robot has been widely studied at home and abroad.This kind of intelligent mechanical equipment can enhance people strength or assist human in walking.In an ideal state,this kind of intelligent machine should be able to do the action that the human want to complete in real time.Therefore,it is the premise that this kind of intelligent equipment can quickly and accurately judge the action that the people want to make(also known as human motion intention recognition).Kinematic and mechanical sensors are used to collect human motion information.On this basis,this paper uses machine learning method and prediction method to recognize and predict the five common movements of human lower limbs: walking,going upstairs,going downstairs,going uphill and going downhill.At the same time,the feasibility of the proposed scheme is verified by experiments.This research can provide the basis for the control of intelligent lower limb exoskeleton and other equipment,so that they can coordinate the movement with users.Firstly,this paper analyzes the structure and movement mechanism of human lower limbs.On this basis,the human motion data detection scheme is designed.The detection scheme includes three angle sensors,one acceleration sensor and one pressure sensor.The moving average filtering method is used to filter the signal noise.At the same time,the continuous motion data is segmented to obtain the motion signal segment which contains only one gait period.Secondly,the time-domain feature extraction method is used to extract 30 features of signals.In order to reduce the dimension of feature vector,principal component analysis(PCA)is used to process the features.Finally,30 features are compressed into 5features,and the compressed features contain 88.22% of the information of the original features.Thirdly,support vector machine(SVM)is used to build the action recognition model and the average recognition accuracy is 85.4%.In order to further improve the accuracy of motion recognition,markov chain model(MCM)is used to describe the conversion rule between human gait motion,and the recognition results of SVM are optimized by using this rule.The experimental results show that the average recognition accuracy of the optimized model is 93.6%.Compared with BP neural network and k-nearest neighbor algorithm,the results show that the proposed method is better than the contrast experiment in recognition accuracy and recognition timeliness.Then,the angle signals of the ankle,knee and hip are segmented into signals segments by the corresponding the extreme points.According to the slope of the first and last points of the signal segment,six signal patterns are defined to describe these signal segments,so that the continuous sensor signals are transformed into discrete values to express the human motion states.Finally,the angle signals of ankle,knee and hip joints are analyzed by the multi-dimensional temporal association rule algorithm,and the temporal correlation of corresponding signal patterns about different movements are mined.Then the signal pattern at the next moment is qualitatively predicted through the established association rules,so as to predict the movement of human lower limbs.The experimental results show that the average accuracy rate of motion prediction is 78.3%,and it can predict the subsequent movement within average 92.24 ms.