We outline a preliminary set of criteria that computer vision algorithms have to satisfy, for an efficient use in visual sensor networks (VSNs). We briefly present an idea on how such a network would operate to achieve optimum performance and propose future work for the Ph.D. studies.