Autonomous driving robots (UGVs, Unmanned Ground Vehicles) equipped with visual-optical (VIS) cameras offer a high potential to automatically detect suspicious occurrences and dangerous or threatening situations on patrol. In order to explore this potential, the scene of interest is recorded first on a reference tour representing the 'everything okay' situation. On further patrols changes are detected with respect to the reference in a two step processing scheme. In the first step, an image retrieval is done to find the reference images that are closest to the current camera image on patrol. This is done efficiently based on precalculated image-to-image registrations of the reference by optimizing image overlap in a local reference search (after a global search when that is needed). In the second step, a robust spatio-temporal change detection is performed that widely compensates 3-D parallax according to variations of the camera position. Various results document the performance of the presented approach.
Change detection on UGV patrols with respect to a reference tour using VIS imagery
2015
2015
Conference paper
Electronic Resource
Unknown
British Library Conference Proceedings | 1992
|Sécurité The operations patrols
Online Contents | 1998
|Guidelines for Establishing Freeway Service Patrols
British Library Conference Proceedings | 1996
|SECURITY NEWS - US steps up patrols
Online Contents | 2004
Bicycle patrols: an underutilized resource
Emerald Group Publishing | 2008
|