Perspective Aware Road Obstacle Detection

While road obstacle detection techniques have become increasingly effective, they typically ignore the fact that, in practice, apparent size of obstacles decreases as their distance to vehicle increases. In this paper, we account for by computing a scale map encoding hypothetical object at every image location. We then leverage perspective (i) generate training data injecting onto synthetic objects whose corresponds foreshortening; and (ii) incorporate information decoding part network guide detector. Our results on standard benchmarks show together, these two strategies significantly boost performance, allowing our approach consistently outperform state-of-the-art methods terms instance-level detection.