Fisheye image rectification for efficient large-scale stereo

Stereo vision using fisheye lenses has received greater attention in recent years due to its vast number of applications. As an essential prerequisite for stereo processing, this work proposes a fisheye image rectification method for efficient large-scale stereo. The scene distance was formulated within the spherical model. Reprojecting the original images using transverse Mercator projection enabled us to proof a theorem for which distance computing could be simplified to only a division operation of a constant and the disparity. Experimental results shows that stereo processing using this method could achieve much faster computational speed than existing approaches while maintaining satisfying matching quality.