Research On Recognition And Application Of Hand Gesture Based On Skin Color And SVM

With the development of computer technology, research on gesture recognition has gradually become a central issue. Gesture as an interface with human-computer interaction has simple and humanity features, therefore this non-contact interaction based on vision can meet the new demands of human-computer interaction. But in the real application, gesture is various and ambiguous because of the different environment and different people, combined with hand-shaped complex, increasing the difficulty by visual identification. So it is still a challenge to achieve the task of gesture recognition without contact.This paper aims at gesture recognition based on vision, gesture detection and recognition based on skin color and SVM, and then uses simple gestures to control the robot in real time. The research content mainly includes these five aspects: gesture image preprocessing, gesture segmentation, gestures features extraction, gesture recognition and robot control simulation.Firstly, image preprocessing. In the aspect of image enhancement, this paper selects the median filter and the laplasse algorithm for smoothing and sharpening of the image, remove noise while keeping image details better. In terms of gesture segmentation, it makes the skin color space from the RGB space to YCrCb space, the light changes interference is avoided. This paper uses the Otsu algorithm to find out the adaptive threshold, and binarization image, separating the skin color region, then through the gesture area criterion eliminates small area of approximate skin color and the face region, so the gesture is extracted from complex background. In the aspect of gesture integrity, we use morphological processing to effective fill "holes" and remove the burr.Then, extracting feature vectors of hand gesture. Bying Detecting and drawing gesture edge profile by calling OpenCV library function. We extract 7-step Hu moments eigenvalues on the gesture contour, then detect fingertip number by the way of calculating the contour curvature and convex hull, that constitutes gesture feature vector with 7-step Hu moments. This paper extracts features through gesture contour with small calculation, and Hu moment feature with pan, zoom and rotation invariance etc.Finally, realizing recognition and application of hand gesture. After extracting the gesture image feature, we construct fearture database. Next step is to train for generate models to classify and recognize hand gesture. On the application, the gesture recognition results is used in robot controling in webots simulation environment. The gesture will be conveyed to the NAO robot and make the corresponding action. It will make real-time control function for robot by gesture recognition.