Gesture Recognition Based on Pixel Classification and Contour Extraction

Most of the existing gesture recognition methods are based on the detecting of fingertips and palm points. Thus we address the techniques of gesture recognition by combining the features of fingertips, wrist and arm. In our method, firstly, rough hand region is segmented from depth image, and then precise hand region is refined by Bayesian skin color model. Secondly, finger and fingertip region is extracted in the segmented precise hand region by the pixel classification. Finally, fingertip points are accurately obtained by the method of ellipse fitting. Contour is drawn at the precise hand region, wrist point is detected by the minimum inscribed rectangle and arm point is detected through inscribed circle of the contour. Compared with other algorithms, the algorithms proposed in this paper can exactly detect feature points for gesture recognition.