A compact, lightweight and energy efficient system for autonomous navigation based on 3D vision

This paper proposes a compact and lightweight system for 3D vision based autonomous navigation. Our navigation system consists of two main components: a stereo vision camera with FPGA processing developed by our research group and an embedded quad core ARM board. These two components allow us to implement a robust obstacle detection algorithm that enables control of a moving vehicle in an unknown (indoor or outdoor) environment. Compared to other vision based navigation system aimed at dealing with this task, our proposal is very compact, lightweight and, thank to the reduced energy requirements, suited for battery powered vehicles. The 3D sensing capabilities provided by the embedded stereo camera deployed make the system very robust. We extensively tested our proposed architecture and navigation system using a small and battery powered rover in very challenging environments. In this paper we provide experimental results concerned with the vision module on sequences acquired in real indoor and outdoor application scenarios. The proposed system is an ideal platform for future developments aimed at 3D registration, visual odometry and object categorization/recognition