Research On Action Recognition Based On Inertial Sensors

With the development of electronic technology and sensor technology,more and more smart electronic devices integrate miniature inertial sensors,which makes the study of human action recognition based on inertial sensing data has great application value,such as human-computer interaction,medical and health,sports and entertainment etc.Motion recognition based on inertial data is a new research direction in the field of pattern recognition.In essence,it is a process of action data acquisition,feature extraction,classification and recognition.Inertial motion information,including acceleration and angular velocity information,is ubiquitous in daily life.Compared with visual information based action recognition,it can directly reflect action meaning and has lower requirement for use environment.Inertial data acquisition is the first step of action recognition,which plays a vital role in the recognition of the results.In this study,MPU6050 inertial sensors are selected for action data acquisition,and Calman digital filtering is used to eliminate noise effects.The minimum normalization method is adopted to unify the numerical range of acceleration and angular velocity.A differential threshold detection method is designed to intercept the inertial data during motion.For continuous actions,cross validation is used to find the optimal window size and intercept the inertia data.Feature extraction is to find out the characteristics that can distinguish different movements most.There are different data features in different fields,and the optimal feature selection is a process of optimization.This paper firstly extracted 48 dimensional motion characteristics,and then finds the optimal feature subset by combining the PCA method with the exhaustive method,the cross validation shows that the recognition rate can be improved effectively.The classification algorithm categorizes the sample to be measured through the feature data,which directly affects the quality of the recognition result.In this paper,the Support Vector Machine algorithm is selected as the data classification algorithm.By designing appropriate kernel functions,the computational complexity of the classification process is reduced.For multi class problems,the hierarchical structure of the directed acyclic graph is optimized to improve the recognition rate.The experiments on six movements,such as walking,running,jumping,upstairs,downstairs and squatting,showed that the recognition rate was over 96%.