HASLOC: Hierarchical And-Or Structures forLocalization and Object Recognition

Target detection and recognition is a challenging problem because of changes in appearance, viewing direction, occlusion and other covariates. Systems that can accurately and efficiently detect and track objects can provide several benefits in surveillance, monitoring and other applications. As part of this effort, we propose to develop a robust learning-based approach to detect, track and recognize targets. Our approach will use a multi-layered, hierarchical model that can handle changes in target's signature and apparent view. Initially the effort will focus on EO/IR image sequences and extend the approach to multispectral data in Phase II.