Person Re-identification Using Overhead View Fisheye Lens Cameras

Person Re-identification has been a widely explored topic in automated video-based surveillance. Nonetheless, the substantial body of work in this area deals only with upright images of persons captured using rectilinear lenses. However, for surveillance applications, it is often preferable to use fisheye lenses to expand the field of view. Person re-identification using fisheye lenses is an uncharted territory with potential applications in image and video-based surveillance. In this paper, we explore various techniques by which we can use images and videos captured by fisheye lenses for person re-identification using conventional algorithms developed for rectilinear images. We discuss the application of foreground extraction and propose a framework for orientation normalization for use in fisheye images. We also introduce a novel radial co-location search to improve the accuracy of person re-identification using fisheye lenses.