Abstract

The problem of Simultaneous Localization And Mapping (SLAM)  originally arose from the robotics community and is closely related to the  problems of camera motion estimation and structure recovery in computer  vision. Recent work in the vision community addressed the SLAM problem  using either active stereo or a single passive camera. The precision of camera  based SLAM was tested in indoor static environments. However the extended  Kalman filters (EKF) as used in these tests are highly sensitive to outliers. For  example, even a single mismatch of some feature point could lead to  catastrophic collapse in both motion and structure estimates. In this paper we  employ a robust-statistics-based condensation approach to the camera motion  estimation problem. The condensation framework maintains multiple motion  hypotheses when ambiguities exist. Employing robust distance functions in the  condensation measurement stage enables  the algorithm  to discard a  considerable fraction of outliers in the data. The experimental results  demonstrate the accuracy and robustness of the proposed method.