A method of optical flow computation based on LUV color space

For improving accuracy of optical flow computation, we propose a new method to compute optical flow from both gray and color information. First, the Gauss filter is used to pre-filter the color image sequence. Then optical flow recovery is based on the method of neighborhood least squares using brightness information. And the weight coefficients of local neighborhood are determined by U and V channels in LUV color space. Finally, the optical flow result is filtered by median filter. The experiment results show that the proposed method produces dense optical flow field and accuracy of the results can also be improved.