Large-Scale Real-Time Object Identification Based on Analytic Features

Inspired by biological findings, we present a system that is able to robustly identify a large number of pre-trained objects in real-time. In contrast to related work, we do not restrict the objects' pose to characteristic views but rotate them freely in hand in front of a cluttered background. We describe the essential system's ingredients, like prototype-based figure-ground segmentation, extraction of brain-like analytic features, and a simple classifier on top. Finally we analyze the performance of the system using databases of varying difficulty.