Control of perceptual attention in robot driving

摘要

Computer vision research aimed at performing general scene understanding has proven to be conceptually difficult and computationally complex. Active vision is a promising approach to solving this problem. Active vision systems use optimized sensor settings, reduced fields of view, and relatively simple algorithms to efficiently extract specific information from a scene. This approach is only appropriate in the context of a task that motivates the selection of the information to extract. While there has been a fair amount of research that describes the extraction processes, there has been little work that investigates how active vision could be used for a realistic task in a dynamic domain. We are studying such a task: driving an autonomous vehicle in traffic.