| The natural interaction based on hand gesture recognition has a wide range of application, which is one of the main human-computer interactions. The Vision-based gesture recognition in this paper includes gesture segmentation, gesture feature extraction, static gesture recognition, and dynamic gesture recognition. The gesture feature extraction is the most important step which determines the accuracy of gesture recognition. Dynamic gesture recognition is to recognize the gesture sequence in order to understand the intentions of human interactions.Visual gesture recognition system runs in the natural environment, which is affected by light. In order to reduce the impact of the light, this paper fuses the background subtraction with the color extraction method based on YCbCr color space to obtain the accurate and complete gesture binary image.Gesture feature extraction consists of two parts, one is the multi-feature fusion of Zernike moments and structural features to recognize the hand shape, and the other is feature extraction of the main direction discrimination for hand gesture through the fitting ellipse. The principal direction is very different from the thumb direction, so the location of fingertip and convex defect depth point are used to remove the impaction of the thumb. Because the features of main direction are extracted after the hand shape feature extraction, so the removal of thumb has no effect on the hand shape extraction. After that, the model match method is used to recognize the static hand gesture.This paper uses dynamic time warping(DTW) algorithm to complete the dynamic gesture recognition. Nine kinds of gestures are classified and recognized which can meet the real-time and the recognition rate both.This method runs by VS2010 and OPENCV. The database adopts the dynamic hand gesture data from University of Cambridge. The experiments show that the average recognition rate of hand gesture is up to 96.85% and the main direction of gesture is up to 99%. |
PDF Full Text Request