Omnidirectional Object Duplicate Detection

In this paper, we extend a graph-based approach for omnidirectional object duplicate detection in still images. Objects are detected from several points of view with different distances. The goal of this work is to determine how many training images have to be taken and from which points of view in order to achieve a certain efficiency. Moreover, the performance of the algorithm is improved by automatically generated images, where the original training images are scaled and rotated in 3D space. Our experiments show that four training images are enough for 3D object duplicate detection from a planar view point and ten training images for omnidirectional detection.