S. Meena, K. Palaniappan, G. Seetharaman, and V. B. S. Prasath

Proc. SPIE Conf. Geospatial Informatics, Fusion, and Motion Video Analytics VI, 2016

interactive, segmentation, wami, target selection

PlainText, Bibtex, Google Scholar

Abstract

Tracking a target (i.e. moving vehicle, human being, object, animal) in airborne aerial imagery is a challenging task. Targets are often small in size, target appearance changes with viewpoint including shape distortions, illumination differences and specularities also significantly influence target appearance, targets may be camouflaged, dynamic backgrounds, complex cluttered backgrounds, etc. Occlusions, presence of other moving targets and additional complexities pose further difficulties. This makes fully automatic detection and segmentation of the target a difficult task. Elastic body splines (EBS) belong to the family of splines and have been applied for the task of biomedical image registration. It models the elastic deformation of homogeneous isotropic elastic body subjected to external forces. In this work, we propose to use EBS within an interactive target selection framework which allows the user to select and delineate a complex target by marking only a few pixels. Interactive EBS-based target segmentation is able to perform an accurate segmentation of the target. Our experimental results on wide area motion imagery (WAMI) indicate that the sparse interaction (labeling only 4-6 pixels) by the user leads to good target detection and segmentations in challenging tracking scenarios.