PEVR: Pose Estimation for Vehicle Re-Identification

Re-identification is a challenging task because the available information is partial. This paper presents an approach to tackle vehicle re-identification (Re-id) problem. We focus on pose estimation for vehicles, which is an important module of vehicle Re-id. Person Re-id received huge attention, while vehicle re-id was ignored, but recently the computer vision community have started focusing on this topic and have tried to solve this problem by only using spatiotemporal information while neglecting the driving direction. The proposed technique is using visual features to find poses of the vehicle which helps to find driving directions. Experiments are conducted on publicly available datasets VeRi and CompCars, the proposed approach got excellent results.