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.


    Access

    Download


    Export, share and cite



    Title :

    Change detection on UGV patrols with respect to a reference tour using VIS imagery


    Contributors:

    Conference:

    2015


    Publication date :

    2015



    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    Unknown




    Bicycle patrols

    Smith, N. / Bicycle Federation of Australia | British Library Conference Proceedings | 1992


    Sécurité The operations patrols

    Peyronne, C. | Online Contents | 1998


    Guidelines for Establishing Freeway Service Patrols

    Fenno, D. / Ogden, M. / Morris, M. et al. | British Library Conference Proceedings | 1996