Multi-Target Parametric Active Contours to Support Ontological Domain Representation

In this paper, we present an efficient multi-target active contour approach useful for the support of ontological domain representations. The developed active contours provide both segmentation and tracking of multiple non-rigid objects evolving over video scenes captured by mobile cameras with unknown calibration properties. The proposed multi-target parametric active contour technique extracts information such as shape, position, but also intensity or color of the detected objects and parts of them. These visual and numeric data are necessary to feed the representation of ontological domains such as this of the spatio-temporal visual ontology which involves semantic concepts of scene objects and sub-objects well as a set of spatio-temporal, geometric and photometric relations between them. Our approach applied on real-world crowded scenes has shown excellent results in detection of semantically related objects of interest.