Fusion of range and vision for real-time motion estimation

Abstract

We introduce a motion estimation algorithm that fuses visual and range data to give an unambiguous estimate of the velocity of objects visible to a camera and range sensor. Dynamic scale space is used to avoid temporal aliasing and a novel robust estimator based on Least Trimmed Squares is used to smooth results between boundaries established using range data. Simulation results (from a specially developed simulation environment) and experimental results (from an FPGA based implementation of our algorithm) show that our approach gives accurate motion estimates.