An Improved Cross-Camera Vehicle Tracking Method: Re-Identification Feature Matching of Confidence Based on Spatio-Temporal Information

Intelligent Vehicle Infrastructure Cooperative Systems is a key research topic in the field of Intelligent Transportation Systems; traffic object perception based on cameras is one of its foundations. Due to the development of computer vision, single-camera traffic object tracking has implemented some advanced methods, but it is still inadequate in identity matching, especially because of more similar appearance. This study uses the background of multi-cameras and the DeepSORT algorithm as the basic framework to propose a vehicle identity matching algorithm based on the re-identification features determined by spatio-temporal information. The proposed method has been tested on a benchmark dataset of traffic video, achieving great performance and verifying its validity. Finally, this paper further discusses the advantages and disadvantages of the cross-camera vehicle tracking algorithm based on joint target matching of vehicle features and spatio-temporal information, contributing to the direction of future improvement of the algorithm.