Image segmentation from motion information

Two schemes for segmentation from an image sequence are presented. The first scheme makes use of the fact that the second temporal derivative of the intensity function can be regarded as a motion weighted spatial Laplacian and only those parts of the image with motion have a nonzero value of the first derivative. However, this method is only applicable to single moving object situations. The second scheme applies an additional constraint. By use of this additional constraint, the optical flow field can be estimated without loss of the sharpness of the moving edges. From this optical flow field, Gaussian-Laplacian operator is applied and the zero-crossings as the moving boundaries or occluding boundaries are found. By this approach, one can easily separate several moving objects with different velocities.<
>