A method for object localization in a multiview multimodal camera system

The aim of the paper is to show that localization of observed moving objects is possible in a multi-camera multimodal environment. The proposed method based on the simple foreground mask and there is no need for further pre-processing steps. The geometrical information for multiview localization also computed by using the statistical evaluation of foreground mask: homography, vertical vanishing point and horizon line are estimated. The method is flexible, it can handle objects of any size in images of different modality. It is based only on change detection masks and does not depend on appearance information. Due to its statistical nature the proposed method efficiently handles such challenging situations as changes in viewpoint, occlusions due to view variations, background clutter. Experimental results validate the usefulness in real-life outdoor environment.