Georeferenced feature tracking in wide field images

In this paper we implemented a system of fisheye vision (a 360° × 180° visual field) for georeferenced tracking of feature points associated with moving objects in urban scenes. To develop this system: 1. We calibrate the intrinsic parameters of a fisheye camera based on the stereographic model. 2. We propose a geometric model of image formation and adjust the position and orientation parameters using satellite images of the observation zone. With this calibration process, we establish a mapping that allows us to convert positions of feature points from the fisheye image to the georeferenced image. 3. We study the uncertainties of the camera pose and the mapping. 4. We propose a system based on SURF and FLANN to detect feature's trajectories in fisheye video. 5. We calculate georeferenced trajectories of vehicles in the fisheye video.