Gesture Recognition Based On Sparse Representation

Gesture recognition is one of important prerequisite of human-computer interaction which complies with the trend of development this technology and plays an important role and significance for human-computer interaction. It becomes a research hotspot in recent years that sparse representation can use less signal represents the target information. In this paper, the sparse representation theory is applied to the field of gesture recognition. The main work of this paper is summarized as follows:Firstly, depth theory of gesture recognition is studied and analyzed. Refer to the feature extraction and description of the human face recognition, the descriptors like Principal Component Analysis (PCA), Speeded Up Robust Features (SURF) are used to extract features of hand images and achieve gesture feature extraction experiments.Secondly, Sparse representation is proposed for the classic gesture recognition method under an angle rotation and offset variation the recognition rate is not high. The route is as follows:first of all, the pre-processing of the training sample of gesture images is being introduced which including the image size adjustment, the training sample size normalization, and transformation of color space. Then we use the feature extraction of PCA method for extracting the training sample and choose sparse representation to classify the training sample. Sparse representation is solved by the least square method, then the redundant dictionary is gained from the Sebastien training samples, and will be used to sparsely represent the test gestures’to classify the gesture images by the residual error minimum classification. The experimental results show that the test signal samples were classified by sparse representation could be able to identify with a higher rate than the classic neighbor nearest classification even if a certain angle rotation and offset variation of the gesture.Thirdly, based on static gesture recognition, further visual-based dynamic gesture recognition study is presented and realized. Dynamic video sequences gesture tracking using Kalman filtering, Gaussian mixture model and Camshift algorithm is achieved respectively and the experimental results were analyzed which has been done for the subsequent dynamic gesture recognition of sparse representation do a previous preliminary exploration research work.