Vehicle Verification Using Deep Learning for Connected Vehicle Sharing Systems

Abstract

Information sharing in connected vehicle systems helps each participating vehicle to have a more complete and expanded sensing range beyond its own sensing capability. When sharing visual traffic information among vehicle nodes, it is of great significance to identify overlapping components and associate objects in common to create an accurate and complete surrounding scene. This paper Extends FusionEye, a study of perception sharing, by exploring deep learning approaches for real time vehicle verification tasks. We propose two deep neural network architectures inspired by ResNet and train the neural networks using FusionEye's dataset. Preliminary results show that when learning from vehicle's appearances and kinematic information, the verification accuracy reaches $92%$, which provides possible solution for real time system.