Vehicle Re-identification Using Convolutional Neural Networks

Abstract

Vehicle re-identification is the process of matching automobiles from one place on the road (one field of vision) to the next. Important traffic characteristics like the trip duration, travel time variability, section density, and partial dynamic origin/destination needs may be acquired by performing vehicle re-identification. However, doing so without using number plates has become challenging since cars experience substantial variations in attitude, angle of view, light, and other factors, all of which have a major influence on vehicle identification performance. To increase each model’s representation ability as much as feasible, we apply a variety of strategies that will bring a major change like using filter grafting, semi-supervised learning, and multi-loss. The tests presented in this paper show that such strategies are successful in addressing challenges within this space. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.