Perspective Aware Road Obstacle Detection

Few road obstacle detection techniques explicitly account for the fact that the apparent size ?of obstacles decreases as their distance to the vehicle increases. In this paper to appear in RA-L, we explicit do so by computing a scale map encoding the apparent size of a hypothetical object at every image location and using at every step of the processing as shown in the figure above.

Our results on standard benchmarks show that this significantly boosts the obstacle detection performance, allowing our approach to consistently outperform state-of-the-art methods in terms of instance-level detection.