AUTHOR=Sun Xiaoming , Chen Yan , Duan Yan , Wang Yongliang , Zhang Junkai , Su Bochao , Li Li 

TITLE=Vehicle re-identification method based on multi-attribute dense linking network combined with distance control module

JOURNAL=Frontiers in Neurorobotics

VOLUME=Volume 17 - 2023

YEAR=2024

URL=https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2023.1294211

DOI=10.3389/fnbot.2023.1294211

ISSN=1662-5218

ABSTRACT=Vehicle re-identification is one of the most important tasks in the intelligent transportation system, but there are still many-problems that have not been solved. The first problem is that the efficiency of vehicle re-identification is low and it takes a long time to complete the identification from a large amount of data. The second problem is that the different images of the same vehicle are very different due to the shooting angle, light and image capturing devices, resulting in low accuracy. To solve the above problems, this paper proposes a vehicle classification method based on a dense connection of multiple attributes and distance control module that combines the color attribute and the category attribute of the vehicle based on the HSV color space. It combines the color attribute and the type attribute of the vehicle, improves the accuracy and combines the multi-attribute dense connection method and improves the classification rate. This algorithm is tested on the VeRi776 dataset and the VeRi-Wild dataset (VeRi-Wild (3000) and VeRi-Wild (5000)). The experimental results show that the computational speed and accuracy are 94.24% and 94%, respectively, and the average precision (mAP) is 0.18%, 1.04% and 2.69% higher than the second algorithm (CTCAL).