BIFOCAL STEREO FOR MULTIPATH PERSON RE-IDENTIFICATION
- 1Robert Bosch, Computer Vision Lab, Hildesheim, Germany
- 2Institute of Photogrammetry and GeoInformation, Leibniz Universität Hannover, Germany
Keywords: Person Re-Identification, PRID, Bifocal Stereo, Multipath Person Re-Identification
Abstract. This work presents an approach for the task of person re-identification by exploiting bifocal stereo cameras. Present monocular person re-identification approaches show a decreasing working distance, when increasing the image resolution to obtain a higher reidentification performance. We propose a novel 3D multipath bifocal approach, containing a rectilinear lens with larger focal length for long range distances and a fish eye lens of a smaller focal length for the near range. The person re-identification performance is at least on par with 2D re-identification approaches but the working distance of the approach is increased and on average 10% more re-identification performance can be achieved in the overlapping field of view compared to a single camera. In addition, the 3D information is exploited from the overlapping field of view to solve potential 2D ambiguities.