Optical Flow Detection Using a General Noise Model for Gradient Constraint

In the usual optical flow detection, the gradient constraint, which expresses the relationship between the gradient of the image intensity and its motion, is combined with the least-squares criterion. From a statistical point of view, this means assuming that only the time derivative of the image intensity contains Gaussian noise. However, it is more reasonable to assume that all the derivatives are observed with Gaussian noise because they are equally computed from pixels containing noise and approximated by finite difference. In this paper, we study a new optical flow detection method based on the latter assumption. Since this method requires the knowledge about the covariance matrix of the noise, we also discuss a method for its estimation. Our experiments show that the proposed method can compute optical flow more accurately than the conventional method.