Enhancing COCOA framework for tracking multiple objects in the presence of occlusion in aerial image sequences

Abstract

Multi object tracking in aerial image sequences is a topic of utmost importance in the field of computer vision for both military and civilian applications. In order to extract valid information about moving targets, it is required to detect and track these targets precisely in the input image sequences. Occlusion is one of the prominent problem areas that hinder efficient object tracking. Spatial reasoning literature fails to distinguish various analyses that are prominent to computer vision. However, recognizing valid occlusion states and mining their transition sequences help in analyzing the pose and motion of multiple interacting objects in the scene. In this paper, we propose an enhancement incorporating occlusion in the existing COCOA framework for tracking in aerial image sequence. The contribution of the paper is the novel idea of extracting occlusion cues as a pre-processing step to aid the tracker. We describe approaches to extract occlusion information in the scene and use it as a cue for efficient tracking. © 2013 Springer.