Realizing active vision by a mobile robot

Abstract

Most active systems use a camera mounted on a manipulator, because the control of the camera motion is easy and accurate. A limited range of the camera motion by the fixed manipulator, however, often prevents the method from real applications. This paper explores a more general method that a camera on a mobile robot freely moves in the environment and acquires the spatial information based on the active vision paradigm. The camera motion is controlled so as to fix its gaze upon a feature point and to keep the distance to the fixation point constant by visual feedback. The camera motion is determined by its rotation estimated from motions of vanishing points of horizontal lines in the environment. Non-horizontal lines are detected by their motion parallax caused by the linear motion. Experimental results show this feedback can control the camera motion within small deviations from the planned path. The deviations measured in the image are used to compensate for the errors in the 3-D position and the camera motion.

Introduction: Machine vision research at Osaka University

“…Li et al [42] controlled robot motion (camera is fixed) to fixate to a target by using vanishing points of the horizontal lines which are used to estimate the camera azimuth Barth, Ishiguro, and Tsuji [41] analyzed camera motion on a mobile robot along the circular path in order to fixate on a target object.…”