High Precision Dense Depth Map Acquisition by an Omnidirectional Stereo System

J.-i. Takiguchi, J.-I. Eino, and T. Hashizume (Japan)


-- Environmental Recognition, Image Processing, Stereo Vision, VRML, ALR, Omnidirectional Stereo System


-- This paper presents the estimation of depth based on the cylindrical re-projections of images captured by a stereo setup consisting of two sets of an ODV (OmniDirectional Vision system). In general, images obtained by such a device have a relatively low spatial resolution and have some aberration in comparison to standard camera images and thus, considerable range estimation errors are caused. We propose a stereo setup of an ODV, which features two mirrors with a contrived curvature to minimize blurring influence. We also propose an edgebased intensity interpolation method, a bi-directional subpixel matching procedure, and a discontinuity compensation algorithm to reduce the disparity error. The 3D-image of the complicated outdoor scene written in VRML(Virtual Reality Modeling Language) and comparison with an ALR( Area Laser Radar) reveals that the proposed optics and the proposed algorithms are valid and effective and can improve the quality of the range estimation.

Important Links:

Go Back