Enhanced Vehicle Re-identification for ITS: A Feature Fusion approach using Deep Learning

TL;DR

This paper proposes a vehicle re-identification framework that combines CNN and transformer features to improve accuracy in intelligent transportation systems, demonstrating significant performance gains over individual models.

Contribution

It introduces a novel feature fusion approach combining CNN and transformer models for vehicle re-identification across CCTV cameras.

Findings

Fused model achieves 61.73% mAP, outperforming standalone CNN or transformer models.

The approach effectively handles challenges like occlusion and illumination changes.

Experimental results validate the superiority of the fusion framework.

Abstract

In recent years, the development of robust Intelligent transportation systems (ITS) is tackled across the globe to provide better traffic efficiency by reducing frequent traffic problems. As an application of ITS, vehicle re-identification has gained ample interest in the domain of computer vision and robotics. Convolutional neural network (CNN) based methods are developed to perform vehicle re-identification to address key challenges such as occlusion, illumination change, scale, etc. The advancement of transformers in computer vision has opened an opportunity to explore the re-identification process further to enhance performance. In this paper, a framework is developed to perform the re-identification of vehicles across CCTV cameras. To perform re-identification, the proposed framework fuses the vehicle representation learned using a CNN and a transformer model. The framework is tested on a dataset that contains 81 unique vehicle identities observed across 20 CCTV cameras. From the experiments, the fused vehicle re-identification framework yields an mAP of 61.73% which is significantly better when compared with the standalone CNN or transformer model.