Robust Object Matching for Persistent Tracking with Heterogeneous Features

Abstract

Tracking objects over a long period of time in realistic environments remains a challenging problem for ground and aerial video surveillance. Matching objects and verifying their identities across multiple spatial and temporal gaps proves to be an effective way to extend tracking range. When an object track is lost due to occlusion or other reasons, we need to learn the object signature and use it to confirm the object's identity against a set of active objects when it appears again. In order to deal with poor image quality and large variations in aerial video tracking, we present in this paper a unified framework that employs a heterogeneous collection of features such as lines, points and regions for robust vehicle matching under variations in illumination, aspect and camera poses. Our approach fully utilizes the characteristics of vehicular objects that consist of relatively large textureless areas delimited by line like features, and demonstrates the important usage of heterogeneous features for different stages of vehicle matching. Experiments demonstrate the enhancement in performance of vehicle identification across multiple sightings using the heterogeneous feature set.