3D flow estimation using perspective and parallel projection images

Abstract

AbstractIn this paper, the authors discuss a new method for using perspective projection image sequences and approximated parallel projection images that were generated from multiple perspective projection images to estimate the 3D flow representing the motion of a target in space. This technique uses the light ray information of multiple perspective projection images that were obtained by capturing the target at different positions to generate the approximated parallel projection images, which are projected in multiple ray directions. It then uses the optical flows that were obtained from both the perspective projection image sequence and approximated parallel projection image sequence to estimate the 3D flow of the measurement points. A feature of this technique is that using images that were projected in the ray directions of the measurement points that were established on a perspective projection image as the approximated parallel projection image sequence corresponding to the perspective projection image sequence makes it unnecessary to map pixel values between stereo images, which is required in the conventional technique. The authors verified through experiments that this technique enabled approximated parallel projection images to be generated and the 3D flow to be estimated satisfactorily by using them. © 2006 Wiley Periodicals, Inc. Syst Comp Jpn, 37(7): 47–60, 2006; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20258