School of Engineering
This paper presents and investigates two models for a multi camera configuration with visual saliency capability. Applications in various imaging fields each have a different set of detection parameters and requirements which would result in the necessity of software changes. The visual saliency capability offered by this multi camera model allows generic detection of conspicuous objects be it human or nonhuman based on simple low level features. As multiple cameras are used, an image stitching technique is employed to allow combination of Field-of-View (FoV) from different camera captures to provide a panoramic detection field. The stitching technique is also used to complement the visual saliency model in this work. In the first model, image stitching is applied to individual captures to provide a wider FoV, whereby the visual saliency algorithm would able to operate on a wide area. For the second model, visual saliency is applied to individual captures. Then, the maps are recombined based on a set of stitching parameters to reinforced salient features present in objects at the FoV overlap regions. Simulations of the two models are conducted and demonstrated for performance evaluation.