Advanced vehicle tracking in persistent aerial surveillance video

Abstract

This paper presents a relational graph based approach to track thousands of vehicles from persistent wide area airborne surveillance (WAAS) videos. Due to the low ground sampling distance and low frame rate, vehicles usually have small size and may travel a long distance between consecutive frames, WAAS videos pose great challenges to correct associate existing tracks with targets. In this paper, we explore road structure information to regulate both object based vertex matching and pair-wise edge matching schemes in a relational graph. The proposed relational graph approach then unifies these two matching schemes into a single cost minimization framework to produce a quadratic optimized association result. The experiments on hours of real WAAS videos demonstrate the relational graph matching framework effectively improves vehicle tracking performance in large scale dense traffic scenarios.