Gesture Recognition Algorithm Based Extreme Learning Machine

With the development of computer technology and human-machine interaction,hand gesture has become a popular but challenging topic in the field of machine learning and digital image processing as a natural,intuitive and efficient interactive mode.Artificial neural network has been widely used in many fields because of its good self-learning,adaptive and distributed storage capacity,it also has application in gesture recognition system.But there are some problems in the traditional neural network,such as slow training speed,easily cause the network converging to local minimum,unable to achieve global minimum.And gesture acquisition process has many non-controlling factors restricting the performance of gesture recognition,such as ambient light,background and even shooting angle.So Gesture image feature extraction,data set reduction and gesture classification were investigated in this paper.The main results of this paper are as follows:(1)This paper summarizes the background and achievements of gesture recognition,and analyzes the current development of the technology.(2)Several typical image processing methods are analyzed and summarized,such as smoothing,binary and morphological processing.The quality of the image can be improved after pretreatment,the effectiveinformation of the image is enhanced,and get rid of irrelevant information to ensure accurate and reliable results of subsequent gesture recognition.(3)The frame-difference method is used to extract the features,then the gesture features are represented by a monogenic feature.The main features are extracted by PCA algorithm and the dimension of the data is reduced.Experiments results prove that the fusion of the single feature representation and the transfinite learning machine algorithm improves speed and accuracy of the recognition,and the effectiveness of algorithm is verified.(4)This paper studies the classification of hand gestures by using hierarchical extreme learning machine(H-ELM)as a classifier in hand gesture recognition.Sparse automatic coding and hierarchical training of H-ELM can get the multi-layer sparse representation of the original input,so that the output of the automatic coding is similar to the original input,minimizing the error of reconstruction,and improving recognition accuracy of gesture feature classification.(5)This paper studies the classification of hand gestures by using online sequential extreme learning machine(OS-ELM)as a classifier in hand gesture recognition.OS-ELM can process data individually or batch processing.After the current data is trained,the space will be released in time,therefore,the training time is shortened and the efficiency of gesturerecognition is improved.