Gesture Recognition And Applications Based On Gabor Feature And Sparse Representation Classification

As a main part of human body, hand plays an important role in people's daily life. Hand gesture is the basic way of human communication. With the development of science and technology, the natural and intuitive human-computer interaction systems based on hand gesture have brought great convenience in people's daily life and the gesture recognition technology is becoming one of the hottest research spots in recent years.This thesis mainly realizes gesture recognition system which contains static gesture and posture changed dynamic gestures in monocular visual environment and its application in music player controller. The gesture recognition system includes establishing the gesture image library, image preprocessing, gesture partition, feature extraction and classification.First, as the existing dynamic hand gestures based on trajectory need to identify the starting and ending position and the gesture realizations in a wide range of motion needs tracking, we define and collect eight kinds of dynamic gesture using infrared camera. Collection of dynamic hand gestures are respectively presented from three aspects: gesture types, users, gesture posture. We all give the corresponding gesture image samples.Second, in the gesture division stage, this thesis proposed the partition of the static hand gesture and the dynamic hand gesture based on the frame difference method. While for different kinds of dynamic hand gestures, we use the difference threshold judgment method. The two fold selections which combine threshold selection of hand segmentation and key-frame selection strategy are used to accomplish dynamic hand segmentation and choose the key frames from the gesture image sets. After the key frame selections the dynamic hand gesture image sequence length is normalized which solve the problem that the speeds of gestures change are different due to individual habits. The experiment proves that, the key frame extraction not only improves the dynamic hand gestures recognition but also reduces the operation time of gesture recognition.In the stage of feature extraction and gesture recognition, the algorithm which fusion Gabor wavelet transformation, principal component analysis (PCA) to reduce the Gabor feature dimension, sparse representation classification (SRC) method is applied to classify gestures. It is proved that the recognition of GSRC algorithm in static gestures as high as 98.57% by a large number of experiments, which is better than the algorithm combining PCA with SRC and the Shift descriptor matching algorithm. In dynamic gesture recognition, the recognition rata can reach 95.71% which is higher than the algorithm combining PCA with SRC.Finally, the system based on above gesture recognition structure is proposed to realize the complete control of the music player. Using the gesture recognition proposed and offline training mode, functions of fast forwarding, fast rewinding, pausing, switching to the next track or the previous track, the volume increasing, and the volume reduction can be realized by the predefined gestures recognition.