ISSN 1054-6618, Pattern Recognition and Image Analysis, 2011, Vol. 21, No. 1, pp. 20–29. © Pleiades Publishing, Ltd., 2011. 1. INTRODUCTION In the last years, the wide availability of embedded systems and low-cost camera sensors—together with the developments in wireless communication—has made it possible to conceive sensor-based pervasive intelligent systems centered on image data [1]. Such visual Wireless Sensor Networks (WSNs), employing a great number of low power camera nodes, may support a large class of novel vision-based applications thanks to the great informative power of imaging. For exam- ple, visual WSNs may be used to monitor in real-time the crowd in shopping mall, airports and stadiums. By mining the scene, the network may detect anomalous and potentially dangerous events [2]. Similarly, visual WSNs may be used for environmental monitoring, for the remote control of elderly patients in e-Health [3] and for human gesture analysis in ambient intelligence applications [4]. In brief, the key feature of visual WSNs is the combination of the versatility and independence from a physical infrastructure typical of general WSNs with the richness of information that can be gained through imaging techniques, computer vision and image understanding. In this sense, a visual WSN may be understood as a distributed and collaborative sensor network, able to produce, aggregate and process images in order to mine the observed scene and communicate the relevant information found about it. In particular, each acquisition node is capable to acquire a view of the scene to be mined. Then, in cooperation with each other, the various devices in the network are able to process and aggregate the acquired views in order to predicate something about the scene. A successful design and development of such a sys- tem cannot be achieved without suitable solutions to the involved computer vision problems. Although the computer vision problems may be still decomposed into basic computational tasks (such us feature extrac- tion, object detection and object recognition), in the context of visual WSNs, it is not directly possible to use all the methods that have been developed to solve such tasks and that are already available in the specific liter- ature [5]. Indeed, since WSNs usually require a large number of sensors, possibly scattered over a large area, the unit cost of each device should be small to make the technology affordable. Therefore, cost constraints limit the computational and transmission power of sensor nodes as well as the fidelity of acquired images, Visual Sensor Networks for Infomobility 1 M. Magrini a , D. Moroni a , C. Nastasi b , P. Pagano b , M. Petracca b , G. Pieri a , C. Salvadori b , and O. Salvetti a a Institute of Information Science and Technologies (ISTI), Italian National Research Council (CNR), Pisa, Italy b ReTis Laboratory (CEIICP), Scuola Superiore Sant’Anna, Pisa, Italy e-mail: massimo.magrini@isti.cnr.it, davide.moroni@isti.cnr.it, christian.nastasi@sssup.it, paolo.pagano@sssup.it, matteo.petracca@sssup.it, gabriele.pieri@isti.cnr.it, claudio.salvadori@assup.it, ovidio.salvetti@isti.cnr.it Abstract—The wide availability of embedded sensor platforms and low-cost cameras—together with the developments in wireless communication—make it now possible the conception of pervasive intelligent sys- tems based on vision. Such systems may be understood as distributed and collaborative sensor networks, able to produce, aggregate and process images in order to understand the observed scene and communicate the relevant information found about it. In this paper, we investigate the peculiarities of visual sensor networks with respect to standard vision systems and we identify possible strategies to accomplish image processing and analysis tasks over them. Although the rather strong constraints in computational and transmission power of embedded platforms that may prevent the use of state of the art computer vision and pattern recognition methods, we argue that multi-node processing methods may be envisaged to decompose a complex task into a hierarchy of computationally simpler problems to be solved over the nodes of the network. These ideas are illustrated by describing an application of visual sensor network to infomobility. In particular, we consider an experimental setting in which several views of a parking lot are acquired by the sensor nodes in the network. By integrating the various views, the network is capable to provide a description of the scene in terms of the available spaces in the parking lot. Keywords: image mining, sensor networks, infomobility, object detection, change detection. DOI: 10.1134/S1054661811010093 Received September 1, 2010 1 The article is published in the original. PROCEEDINGS OF IMTA-3-2010