| The quality of data returned by a sensor network is a crucial parameter of performance since it governs the range of applications that are feasible to be developed using that network. Higher resolution data, in most situations, enables more applications and improves the reliability of existing ones. In this thesis we discuss methods that use controlled motion to increase the image resolution in a network of cameras. In our prototype system, our methods can provide up to 15000x advantage in resolution, depending on tolerable trade-offs in sensing delay.; Mobility itself may have a high resource overhead, and hence a constrained form of mobility is exploited, which has low overheads but provides significant reconfiguration potential. Specifically, we concentrate on pan, tilt, and zoom motion for cameras. Other forms of constrained motion are also mentioned.; An architecture that allows each node in the network to learn the medium and phenomenon characteristics is presented. A quantitative metric for sensing performance is defined based on real sensor and medium characteristics. The problem of determining the desirable network configuration is expressed as an optimization of this metric. A distributed optimization algorithm is developed to compute a desirable network configuration and adapt it to environmental changes.; A key property of our algorithm is that convergence to a desirable configuration can be proved even though no global coordination is involved. Other desirable properties of the algorithm, such as convergence accuracy and convergence time are also studied. Its relationship to previously known optimization heuristics is also discussed.; A network protocol to implement this algorithm is discussed for execution in a totally distributed manner. The protocol involves exchanging messages only in a well-defined neighborhood.; We evaluate our methods using simulations and experiments on our prototype system. Real world data is used for testing the algorithm. |
