Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Computation of discontinuous optical flow by domain decomposition and shape optimization

: Schnörr, C.

British Machine Vision Association -BMVA-:
BMVC 90. Proceedings
Oxford, 1990
S.109-114 : Abb.,Lit.
British Machine Vision Association (Conference) <1990, Oxford>
Fraunhofer IITB ( IOSB) ()
computational vision; image sequence analysis; motion boundaries; optical flow; segmentation

We introduce an approach for the identification and tracking image regions in monocular image sequences. The assumptions are that the egomotion is known and that an representation of the environment has been computed in order to predict roughly the expected motion field with respect to the movement of the sensor relative to the environment. Image regions, whose motions do not match with this motion field, are assumed to belong to independently moving objects in the scene. Having deteced roughly the location and movement of such a region in the image plane (by Optical Flow or any other approach), we decompose the image plane in two disjoint domains and after their shape such that the velocity fiedls which are computed for both domains agree in an optimal way our interpretation of the scene. By this method we reconstruct a velocity field for the whole domain of the image plane, which is discontinuous along a closed contour. This contour coincides an optimal just described way with the mo tion boundaries of the object. As a result, a segmentation of the velocity fields a posteriori becomes superfluous.