Robust Structure from Motion and Identified Dynamics.pdf

Robust Structure from Motion and Identified Dynamics∗ Benjamin R. Fransen2 Octavia I. Camps1,2 Mario Sznaier1 1 Dept. of Electrical Engineering 2 Dept. of Computer Science and Engineering The Pennsylvania State University University Park, PA 16802 Abstract paper, reintroducing these predictions into a Sum-Square- Difference (SSD) based tracker leads to substantial tracking This paper addresses robust recovery of structure and mo- error reduction. Similarly, object appearance errors are re- tion for rigid bodies in video sequences. For small inter- duced by combining the CF identified motion with a piece frame motion, feature appearance is commonly estimated wise planar object model obtained from the sparsely tracked according to the previous frame. However, in the case of features. occlusion or corrupt frames, the small inter-frame motion model fails. In this paper, we propose to use robust identi- fication techniques to estimate the motion dynamics based on a set of previous frames. These dynamics are then recur- 2 Feature Tracking sively used to robustly estimate object structure and motion and to predict the object appearance in future frames. Re- sults are tested for 3d reconstructions of rigid bodies in real At a fundamental level, feature based structure from mo- and syn