FPGA implementation of a vision-based motion estimation algorithm for an underwater robot

Abstract

This paper presents an FPGA implementation for real-time motion estimation of an underwater robot using computer vision. The algorithm searches for correspondences of a given number of interest points for every image acquired by the camera and some previous reference images. In order to minimise the lighting problems, normalised correlation is used as similarity measurement to match corresponding points in different images. The complexity of normalised correlation criteria determined two main parts in our hardware implementation: an array of Processing Elements (PE) and Post Processing Element (PPE).