26 June 1997 Using Lie group method image motion and stereo analysis algorithms for obstacle perception
Author Affiliations +
Abstract
Obstacles to the ground vehicles on the road sometimes is very difficult to detect by active sensors. For example, a fallen tree on the road may not be higher than a speed bump. It is known that human vision system is very good at perception of depth discontinuity and surface orientation change, particularly deep concavity, while comparatively weak at absolute measurement of 3D distance. This is just opposite to the active sensor which are accurate at absolute measurements at isolated surface points, but lack of direct or indirect sensing mechanism of these tow aspects of scene. The obstacle avoidance is more relied on the detection of occlusion edges than that of absolute depth structure. In the past we have modeled the early vision process of primate's visual cortex for binocular and motion based stereo vision using Lie group theory and implemented the algorithm in computers. The algorithm provides surface orientation and range information at places in a scene where the surface regular, namely, locally flat and our energy minimization scheme converges. In a real scene there are plenty of places where the surface structure are not regular, and the energy minimization scheme does not converge. The 'high energy' location will mark the occlusion edge. A further test of 'structure lost' will confirm the occlusion edge, and thus provide the essential information for occlusion avoidance.
© (1997) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Thomas R. Tsao, Thomas R. Tsao, } "Using Lie group method image motion and stereo analysis algorithms for obstacle perception", Proc. SPIE 3087, Navigation and Control Technologies for Unmanned Systems II, (26 June 1997); doi: 10.1117/12.277212; https://doi.org/10.1117/12.277212
PROCEEDINGS
8 PAGES


SHARE
Back to Top