Abstract
In this paper, we present an adaptive and accurate motion estimation algorithm for computing optical flow from an image sequence. The proposed algorithm is based on a regularization formulation that minimizes a combination of modified data-constraint energy and a smoothness measure all over the pixels in the image domain. Unlike the conventional gradient-based optical flow computation, the data constraint used in this paper is derived from the conservation of the Laplacian-of-Gaussian (LoG) filtered image intensity in the temporal sequence. This modification alleviates the problem with the brightness constancy assumption under non-uniform illumination variations due to the use of Laplacian-of-Gaussian filter to remove the low-frequency portion of the additive non-uniform illumination factor. The modified data constraint simply replaces the image intensity function in the image flow constraint by the LoG filtered intensity function. In addition, an adaptively weighted membrane smoothness constraint is employed in our regularization framework. The resulting energy minimization is accomplished by a incomplete Cholesky preconditioned conjugate gradient algorithm. Experimental results on synthetic and real image sequences for our algorithm are given to demonstrate its performance.