Fast vanishing points detection for omnidirectional images

This paper presents a novel method for extracting the dominant vanishing directions from omnidirectional images using the detected conics and lines. The method assumes the camera system has vertically aligned optical axis and is partially calibrated, i.e. only mirror parameter and FOV are known. Based on the geometric properties of catadioptric line projection, the algorithm fully recovers the other camera calibration parameters, cluster the detected conics and use the conic intersections to derive the vanishing points (VPs). The approach shows applying certain geometric constraints can enable fast conic grouping and VPs detection. The efficiency of our method has been validated by experiments on real images (publicly available)