<title>Fast algorithm for obtaining dense depth maps for high-speed navigation</title>

Abstract

Typical stereo algorithms produce sparse depth maps. The depth points often lie on object boundaries and thus obstacles cannot be distinguished from holes. Surface reconstruction algorithms work poorly with such sparse and irregularly spaced data points. We present an algorithm which uses intensity images and a sparse depth map to produce a depth map which can be used for navigation. Our approach is based on an intensity image segmentation algorithm followed by local surface fitting. The algorithm is fast and, thus, suitable for high speed navigation. Results using laboratoty images are presented.