Robust detection of straight and circular road segments in noisy aerial images

This paper treats the problem of detecting straight or circular pieces of road in noisy low-resolution aerial images. it first uses a local operator to detect pixels whose neighborhoods are line-like, and then applies (robust) estimation techniques to find sets of such pixels that lie on or near straight or circular loci. An (unbiased) ordinary least squares estimator cannot handle outlying data even for straight loci; on the other hand, conventional robust techniques for fitting circular arcs are severely affected by digitization effects and the fact that circular road segments are typically short and shallow. We therefore introduce an estimator that is both robust and statistically efficient. We also present a simple ad hoc technique that achieves comparable results and can handle road segments that are either straight or circular. (C) 1997 Pattern Recognition Society. Published by Elsevier Science Ltd.