MULTIWINDOW LEAST-SQUARES APPROACH TO THE ESTIMATION OF OPTICAL-FLOW WITH DISCONTINUITIES

The use of optical flow fields in image sequence analysis allows us to perform motion-based segmentation as well as 3-D reconstruction. Many techniques for optical flow estimation are based on some global or pi ecewise global smoothness assumption. Other techniques compute the flo w field based only on local information. A local algorithm explicitly addressing the problem of evaluating a reliable optical flow field at motion boundaries is presented. Velocity vectors are computed as solut ions of a multiwindow least-squares problem; the field is then regular ized by a vector median filter. The algorithm is noniterative and nonp arametric. Results on both synthetic and real-world sequences are show n; a performance comparison with two well-known techniques demonstrate s the effectiveness of the algorithm in terms of noise rejection, moti on boundary preservation, and speed.