Aiming at the problem of low gesture recognition rate and poor robustness of vision a gesture recognition method based on YCb Cr skin color model of ellipse clustering segmentation gesture and back propagation（ BP） neural network recognition is proposed. Firstly

using discrete cosine transform（ DCT） for delete noising and edge detection on the captured image sequences. Then extract contour profile of gesture according to the human skin color in the YCb Cr space clustering characteristics

obtain the gestures by combining the results of edge detection with the skin color model segmentation. Extracting feature vectors of gestures by using speeded up robost feature（ SURF） algorithm

and finally classify and recognize hand gestures through the BP neural network.The experimental results show that for gestures under complex background the algorithm has strong robustness

high efficiency and the accuracy of recognition can reach to 96 %.