Object Tracking and Background Estimation with a Moving Camera

Abstract

For navigation of vehicles or robots in an environment where other moving objects may be present, algorithms for detection and tracking of moving objects in image sequences are required that can be applied even if the camera itself is moving. A similar problem arises if the camera is mounted on a pole in an outside environment, e. g. for traffic monitoring, and is subject to mechanical disturbances causing jitter in the image sequence. In both cases a dynamical model for the egomotion of the camera can be used for motion estimation and motion compensation. The resulting procedure combines local image matching, Kalman filtering, autoregressive (AR) parameter estimation, background image estimation, and change detection for tracking the moving objects.