Detection of Multiple Deformable Objects using PCA-SIFT

In this paper, we address the problem of identifying and localizing multiple instances of highly deformable objects in real-time video data. We present an approach which uses PCA-SIFT (Scale Invariant Feature Transform) in combination with a clustered voting scheme to achieve detection and localization of multiple objects while providing robustness against rapid shape deformation, partial occlusion, and perspective changes. We test our approach in two highly deformable robot domains and evaluate its performance using ROC (Receiver Operating Characteristic) statistics.