A Comparison of Regional Feature Detectors in Panoramic Images

We present a novel approach to detect and describe visual features in panoramic image data. For various applications, especially computer and robot vision, robust and invariant features are key paths to explore scenes and objects. Most features applied in the literature can commonly be classified either as being local or being global. Local features characterize a significant point in the image like an edge. Global features describe a general property of the whole image like the color distribution. In this paper, we propose an in-between representation using region-based symmetry features. We compare the approach to a set of state-of-the-art affine feature detectors. Experiments show that the symmetry features are sparse, distinctive and robust to changes in panoramic image warp. Therefore, they are well applicable to robot vision tasks