A Deep Learning Approach for Vehicle Re-Identification

Abstract

Vehicle re-identification is currently one of the most important topics within the scientific community. Registration plate recognition may not be a sufficient solution in environments where the vehicle is observed from particular angles (e.g, laterally), or in low light conditions. For this reason, in this paper, we will focus on the study of alternative solutions, able to extract information useful for re-identification regardless of the license plate. Approaches based on the Convolutional Neural Network (CNN) will be analyzed in order to implement a methodology that can learn the salient characteristics of a vehicle and exploit them for the re-identification of the same in different areas, or at different times. The proposed approach will be tested on VeRI 776 datasets and it will be demonstrated that it overcomes the state-of-the-art approaches.