People Tracking in Panoramic Video for Guiding Robots

A guiding robot aims to effectively bring people and from specific places within environments that are possibly unknown them. During this operation the should be able detect track accompanied person, trying never lose sight of her/him. solution minimize event is use an omnidirectional camera: its $$360^{\circ }$$ Field View (FoV) guarantees any framed object cannot leave FoV if not occluded or very far sensor. However, acquired panoramic videos introduce new challenges in perception tasks such as detection tracking, including large size images processed, distortion effects introduced by cylindrical projection periodic nature images. In paper, we propose a set targeted methods allow adapt standard tracking pipeline originally designed for perspective cameras. Our have been implemented tested inside deep learning-based framework with commercial camera. Experiments performed on datasets specifically applications real service show effectiveness proposed approach over other state-of-the-art systems. We release paper annotated open-source implementation our method.