Algorithm And Application Research Of Gesture Recognition Based On SEMG

With the development of robot technology,the new type of human-computer interaction is breaking through the limits of the traditional media.Rearch turned from the program control of the passive receiver instruction to the active understanding of the intentional behavior of the human,which leads to a series of human-computer interaction,such as speech recognition,face recognition,gesture recognition,augmented reality and other emerging technologies.The gesture recognition based on bioelectricity has been paid more attention in recent years because of its natural intuitionistic information representation function and unrestricted background conditions.It has become a frontier subject of new man-machine interaction.Gesture recognition based on sEMG is an important branch in the field 'of gesture recognition.At present most of the gesture recognition based on visual information,through the acquisition of video or image information to extract its contours,then use different recognition algorithm for gesture recognition,which is easily to be affected by complex background,light,angle.Gesture recognition based on sEMG signal is a more effective way to decode the behavioral information from biological signals,which is a deeper mechanism of recognition.This article mainly includes the following contents:Gesture recognition based on the sEMG:based on the original foundation of off-line analysis,the real-time classification of EMG signals is realized.Firstly,we studied the algorithms of online FIR filtering and propose an improved active segment extraction technique based on multi-threshold detection.Then we extract the features of time domain,frequency domain and time-frequency domain separately,and then compare the advantages and disadvantages of four classifiers.Finally,random forest is selected as the classifier of gesture recognition for motion control of home service robots.The recognition rate is 97.32 ± 1.1%,and the delay time is 135 ± 21.6msc,which meets the actual demand.Sign language recognition based on element information:sign language recognition is a branch of gesture recognition,which is more complex and diverse.From the point of feature information,a tree structure classifier which based on oritation,locationand magnitude is constructed for 120 sign language words.Then classify it through a layer of screening,and finally use the improved GMM-HMM model to separate the isolated words.The experiment shows that this method The accuracy rate than directly through the GMM-HMM recognition of 2%to 7%increase.On the basis of this,a tree search model based on AC automata is proposed to improve the recognition rate and error correction ability of sentences.Finally,the algorithm of sign language recognition is applied to the sign language translation system.The recognition rate of isolated words is about 96%,and the recognition rate of whole sentences is about 83%.In addition,we studied the effect of gesture recognition under different conditions.Including different positions of sensors,different combinations of features and different kernel functions of classifiers.And finally the fault-tolerant problem of classifier in data loss and fatigue state is explored.