3D Position Control of Robots Based on Distributed Real-time Vision Navigation System

Indoor localization and navigation of mobile robots has remained as a challenging problem. Vision-based navigation systems have been utilized as a promising solution. But they come with deficiencies such as weak robustness against camera failures. In this paper a vision-based 3D localization and control method for aerial robots is proposed. This method addresses several key features such as higher precision, real-time processing, full area coverage, reduced camera costs and robustness against asynchronized images and camera failures. Most of the processes of the proposed method, including image preprocessing, object recognition and 2D target localization are carried out separately for each camera. Then the acquired data are fused by a distributed Kalman filter to estimate the 3D location of the robot. The method has been experimentally verified by controlling the indoor position of a small aerial robot.