Deep Highway Multi-Camera Vehicle Re-ID with Tracking Context

Abstract

While object detection and re-identification has become increasingly popular in computer version, The growing explosion in the use of surveillance cameras on highway highlights the importance of intelligent surveillance.multi-camera vehicle Tracking, aiming to seek out all images of vehicle of interest in different cameras, can provide abundant information such as vehicle movement for highway supervision department. This paper focus on a interesting but challenging problem, building a real-time highway vehicle tracking framework. We design a two-stage deep learning-based algorithm framework, including vehicle detection and vehicle re-identification. Vehicle re-identification is the most significant part in this tracking framework, however, the most existing methods for vehicle Re-ID focus on the appearance or texture of single vehicle image and achieve limited performance. In this paper, we propose a novel deep learning-based network named VTC (Vehicle Tracking Context) to extract features from vehicle tracking context. Extensive experimental results demonstrate the effectiveness of our method, furthermore, intelligent surveillance system based on proposed tracking framework has been successfully use in Beijing-Hong Kong-Macao Expressway.