Two-step Kalman process for sensor and object motion estimation

This paper addresses the problem of estimating the sensor ego-motion and the motion of rigid objects in a monocular image sequence. It has been developed for a system processing infrared image sequences. These image sequences suffer from a high amount of noise and clutter. Therefore it is necessary to perform long-term image filtering. Since the sensor and the objects are subject to motion, the image sequences have to be motion compensated before filtering can take place. We present a technique based on the well known extended Kalman filter (EKF). It is adapted to the problem of estimating the sensor ego-motion via a correlation based tracking of the horizon. A general model for estimating rigid object motion with EKFs is developed. Since the performance of the EKf in this case strongly depends on its initialization, we propose a special initialization method using a second modified EKF.© (1994) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.