Tracking in Wide Area Motion Imagery Using Phase Vector Fields

Abstract

Tracking in wide area motion imagery (WAMI) is extremely complex because of low resolution of targets, low frame rate and a host of other detrimental environmental factors. In this paper, we propose a robust feature-based method to track objects in WAMI data by exploiting local phase and orientation information based on the monogenic signal representation. We present a detailed monogenic space based analysis to develop a robust method to track objects of low resolution in wide area aerial surveillance imagery. By exploiting local phase information at multiple scales, objects in shadows can be tracked. We also propose an efficient technique to make the feature representation of objects rotation invariant by utilizing local orientation of the object region. The proposed technique is shown to be robust in challenging situations that are characteristic of wide area motion imagery. Effectiveness of the proposed method is illustrated by comparing its performance with dense set of SIFT features and mean shift tracker.