Research On Digital Gesture Recognition Method Based On Feature Motion Sequence

Dynamic gesture recognition is an important field of artificial intelligence.It has a wide range of application prospects in virtual reality,communication between deaf and dumb people,robot control and so on.In recent years,with the rapid development of integrated technology,the research of gesture recognition based on wearable devices has attracted more and more attention.This paper summarizes the current situation and analyzes the key technical features of gesture recognition based on MEMS,then proposes a gesture recognition method based on feature motion sequence.The proposed gesture recognition method mainly includes data preprocessing,feature motion recognition and gesture recognition based on feature motion.First,in the process of gesture data preprocessing,the effective feature set of gesture recognition is obtained by the feature extension combined with feature selection in view of raw data and features affecting the upper limit recognition of data mining and machine learning.Feature extensions perform linear and nonlinear feature transformations on the original data feature sets to obtain candidate feature sets for gesture recognition.A attribute reduction methods based on neighborhood granulation and inconsistent information metrics are proposed to obtain feature subsets that make gesture recognition accuracy optimal.Experimental results show that the classification accuracy of the obtained feature subsets combined with support vector machines,regression trees and other classification methods on multiple UCI common data sets is significantly improved.Secondly,in the process of feature motion recognition,aiming at the randomness of FCM clustering center which affects the recognition accuracy and recognition stability,multi-center fuzzy c-means algorithm is used to extract feature motion recognition adaptively and unsupervised for reduce the subjective influence when human defined.In gesture recognition based on feature motions,the traditional edit distance algorithm is improved to make the cost of edit can reflect the data similarity better.Based on the principle of maximum similarity,the effects of different clustering algorithms and similarity measurement algorithms on gesture recognition are compared.The experimental results show that the improved algorithm can obviously improve the accuracy and stability of gesture recognition,and the accuracy of gesture recognition can reach 98%Finally,the software and hardware of gesture recognition system,gesture acquisition requirements and part of gesture data are introduced.