Tracking by Omni-directional Overlooking Image of Intersection using Fish-eye Camera

Abstract

In this paper, the authors propose an omni-directional overlooking image of intersection using a fish-eye camera, and describe on the image processing method utilizing a feature of the image. The intersection is shot by one camera at a corner of the intersection. The camera turns just under. In the method, an original image from the fish-eye camera is transformed by a pseudo fish-eye image transformation and a spherical polar transformation. The transformed image expresses a scene which is looking down the omni-directional roads similarly. According to the transformed image, all roads are expressed vertically and all pedestrian crossings are expressed sideways. Utilizing the feature of the image, it is possible to detect the cars and the pedestrians of all roads. It is also possible to obtain a data for judgement of a crossing between the car and the pedestrian or between the cars. In the paper, the method is explained and an experimental result is shown.