Tracking Feature Points of Fisheye Full-View Image by Normalized Image Patch

抄録

This paper proposes a method of fast tracking of feature points in the fisheye full-view image captured from a fisheye full-view image sensor. The fisheye full-view image consists of a pair of fisheye sub-images. Each fisheye sub-image of the full-view image sensor has wider field of view than a hemisphere. The local region of feature points of scenes changes abruptly when the feature points move from one of the pair of fisheye sub-images to the other due to the heavy distortion of fisheye images. In this paper, the tracking of feature points is performed via a spherical model. The similarity of local regions of feature points is measured by the normalized image patch of the local region based on the spherical model. In the tracking process, the Kalman Filter is used to predict the location of feature points based on the spherical model. In the experiment, the processing speed with 14 full-view frames per second is achieved on a PC.