Estimating Road Widths From Remote Sensing Images

In this paper, we propose a learning-based approach for road width estimation from high resolution satellite or aerial images. With the road centerline given, a proposed pixel-wise descriptor describes the distribution of the road-like pixels. Then, in the predictive phase, we demonstrate that performance is significantly improved by considering the spatial coherence of the output labels. Experimental results show that the proposed approach provides high quality width estimation results on roads in various terrains.