Automatic Inference of Road Networks from Aerial Images

Favyen Bastani, Songtao He, Mohammad Alizadeh, Hari Balakrishnan, Samuel Madden, Sanjay Chawla, Sofiane Abbar, David DeWitt
To appear in Computer Vision and Pattern Recognition (CVPR), Salt Lake City, UT, June 2018

The availability of highly accurate maps has become paramount due to the increasing importance of location-based mobile applications as well as autonomous vehicles. However, mapping roads is currently an expensive and human-intensive process. High-resolution aerial imagery provides a promising avenue to automatically infer a road network. Prior work uses convolutional neural networks (CNN) to detect which pixels belong to a road (segmentation), and then uses complex post-processing heuristics to infer graph connectivity. We show that these segmentation methods have high error rates (poor precision) because noisy CNN outputs are difficult to correct. We propose a new method, incremental graph construction (IGC), to construct a road network graph from aerial images. IGC is a search process guided by a CNN-based decision function, and unlike prior work, IGC derives the road network graph dircetly from the output of the CNN. We train the CNN to output the direction of roads traversing a supplied point in the satellite imagery, and then use this CNN to construct the graph. We compare IGC with a segmentation method on fifteen cities, finding that segmentation has an 82% higher error rate than IGC in identifying junctions across these cities (13.7% versus 7.5%).

Bibtex Entry:

   author =       "Favyen Bastani and Songtao He and Mohammad Alizadeh and Hari Balakrishnan and Samuel Madden and Sanjay Chawla and Sofiane Abbar and David DeWitt",
   title =        "{Automatic Inference of Road Networks from Aerial Images}",
   booktitle =    {Computer Vision and Pattern Recognition (CVPR)},
   year =         {2018},
   month =        {June},
   address =      {Salt Lake City, UT}