3D recovery using calibrated active camera

Abstract

We describe a system to generate dense depth maps from image sequences recorded from a camera mounted on a robot. Two color images of the sequence are treated as a stereo pair. A color edge detector is applied, lines are segmented from the edge images, and approximations as sequences of straight lines and circular arcs are computed. An initial depth map results from lines of these compound segmentation objects. This map is refined to a dense map by a block matching algorithm. We describe results of a parallel implementation using object-oriented programming techniques. The results show substantial improvements in comparison to a monochrome system with respect to speed, accuracy, and completeness.