Abstract
In this paper, we propose a highly accurate feature point tracking method for omnidirectional image sequence. It is often difficult to extract and track feature points because of nonlinear distortion of omnidirectional images. We have already proposed a method of tracking feature points after transforming the vicinity of a feature point into perspective projection, but sometimes tracking performance was not accurate. In the proposed method, feature points are extracted from each face of regular icosahedron on which omnidirectional image is projected. Furthermore we introduce ASIFT into the patch, which enables robust tracking of deformation. Performance of the proposed method is evaluated through experiments.
