Road Detection from Satellite Images

In this paper, we propose a new method for detecting roads from satellite images. Existing methods have difficulty detecting roads obscured by objects such as trees. To address this problem, our method first uses semantic segmentation to detect road regions that may be obscured by trees. Next, it uses image inpainting techniques to restore the regions. Finally, it performs road detection on the restored images. Experimental results show that our method achieves higher accuracy than existing baselines.