Automatic target segmentation is complicated by the highly variable appearance of targets in imagery taken under realistic operating conditions. In many situations, high-resolution, high-fidelity imagery only complicates the isolation of targets from non-target background. In this paper we present an analysis of the signatures of military vehicles in low-resolution, low-fidelity passive millimeter-wave (PMMW) imagery. A simple model based on Gaussian curvature is developed that characterizes a wide range of target types. This model is used to segment imagery into binary target/non-target regions. The performance of this algorithm is compared to a set of algorithms operating on co-registered laser radar (LADAR) imagery. The comparison shows the PMMW segmentation algorithm producing much fewer false alarm regions, but missing more targets than the LADAR algorithms. We also demonstrate that PMMW sensors show great promise for target detection and for cueing subsequent target identification algorithms. © 2003 Elsevier Science B.V. All rights reserved.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below