Line Detection by Spherical Gradient

Shigang Li, Hanchao Jia, Isao Nakanishi

Abstract: This paper proposes a novel method of detecting lines from omnidirectional image based on spherical gradient of edge points. Since the spherical gradient of edge points is aligned with the normal vector of the line’s projection plane, a line can be determined directly from the spherical gradient direction of the edge points. Since the spherical gradient of the edge points at the two sides of a bar belongs to different hemispheres, the boundary of a bar can be detected easily even if the bar is very narrow. In this paper, the spherical gradient of edge points is incorporated into the Hough Transform to detect lines from omnidirectional image. The experiment shows that the proposed method outperforms the conventional method in not only the accuracy of line detection of narrow bars, but also the efficiency of computation.