This research work was pursued at Irisa/Temics with Jurgen Stauder. We present two approaches for automatic analysis of moving objects in a video sequence. Our main goal in this project is to estimate the 3D motion of a rigid 3D-object between two instants.
The first method estimates 3D motion by an approach based on epipolar geometry under the assumption the camera can be approximated by orthogonal projection.
The second method estimates 3D motion by an optical flow approach. The orthogonal projection assumption still holds, and we assume also that the 3D shape of the object (or at least a rough convex approximation) is known. Consequently, more motion parameters can be estimated.
Both methods are compared with a third one based on object shading.