Monocular Detection and Estimation of Moving Obstacles for Robot Navigation.

Abstract

Abstract — The detection of motion and moving objects or persons with stationary monocular cameras has been extensively studied. However, those techniques fail if the camera is moving itself. In this paper, we present a method for detecting and estimating the position of moving objects using a monocular camera that is mounted in front of a mobile robot platform. The position estimates are used for obstacle avoidance and robot navigation. We apply image warping to compensate the ego-motion of the camera. This allows us to use standard techniques for motion detection. The final position and velocity estimates are obtained using Extended Kalman Filters. Combined with a monocular scene reconstruction our approach allows the robust detection and avoidance of both static and moving obstacles by using a single monocular camera as the only sensor. Index Terms — visual obstacle detection, motion detection, monocular scene reconstruction I.