Dynamical Systems and Motion Vision.
MASSACHUSETTS INST OF TECH CAMBRIDGE ARTIFICIAL INTELLIGENCE LAB
Pagination or Media Count:
This paper shows how the theory of dynamical systems can be employed to solve problems in motion vision. It develops algorithms for recovery of dense depth maps and motion parameters using state space observers or filters. Start with a review of previous, related work followed by an overview of relevant aspects of the theory of dynamical systems. Three dynamical models of the imaging situation are presented The first model assumes that motion is known and a reflectance model for the surface is given. Depth is recovered directly from brightness measurements with a nonlinear Kalman filter. The second model assumes that a planar surface is being reviewed. Motion and depth are recovered with a nonlinear observer. The third model makes no explicit assumptions about motion or surface. It is embedded in a system that iteratively improves an estimate for both the motion parameters and a dense depth map using Kalman filters. No feature-matching preprocessor is required. Solutions to other motion vision problems tracking, camera parameter estimation using dynamical systems theory are suggested. Keywords Dynamical systems Motion vision Kalman filter Depth maps Motion recovery. jhd