Seeing cheaply: flexible vision for small devices

Abstract. In this paper, we outline an architecture for supporting real time autonomous vision in small devices. Our vision processor performs data reduction at the point of image capture, using re-configurable hardware devices to extract and describe image features. By implementing our design on FPGAs, the vision processor can support highly parallel vision processing structures as well as more traditional pipelined or sequential algorithms. Because the processor hardware can be reconfigured, it becomes a flexible prototyping tool. The resulting platform is well suited for research in reactive robot control based on vision, as well as research into biomorphic vision structures.