Extraction and classification of moving targets in multi-sensory MAMI-1 data collection

Abstract

In this work we consider the problem of extraction and classification of moving targets in wide area imagery. We use the Air Force Research Laboratory's (AFRL) airborne multi-sensor dataset, MAMI-1, for testing, wherein moving targets mostly consist of people and vehicles. The movers are extracted using a novel sparse and low-rank matrix decomposition technique. We further compare the classification performance based on SIFT, Dense SIFT, and a superpixel based feature extraction. The results show the superpixel approach as the most advantageous.