05/11/2004, 15:00 — 16:00 — Room P10, Mathematics Building
André Martins, Instituto de Sistemas e Robótica
Estimating Camera Orientation from Video in a Manhattan World
The problem of inferring the 3D position and orientation of a camera that acquires a video sequence is of most interest in Computer Vision. Most approaches require an intermediate step where image features (edges, corners, etc) are detected at each frame and then corresponded among consecutive frames through tracking algorithms. This step is seen as the main bottleneck of those approaches.
We propose a new 3D orientation estimation method for urban (indoor and outdoor) environments, which avoids correspondences between frames. The basic scene property exploited by our method is that many edges are oriented along three orthogonal directions; this is the recently introduced Manhattan world (MW) assumption. We use the octohedral group of symmetries of a cube to derive equivalence classes for the orientation, thus achieving a considerable reduction in the search space.
In addition to the novel adoption of the MW assumption for video analysis, we introduce the small rotation (SR) assumption, that expresses the fact that the video camera undergoes a smooth 3D motion.
From these assumptions, we build a probabilistic estimation approach and demonstrate the performance of our method using real video sequences.
