Cameras have become ubiquitous in our society. Not only are they built into things of everyday life like phones or cars, they are also used in surveillance of public and private spaces. Smart cameras which are able to process the captured video data onboard can be connected to large networks (visual sensor networks). Typically, visual sensor networks are composed of resource-limited devices. Here, coordination among the devices is needed to enable an efficient use of resources. Especially in cases where visual sensor networks are deployed without network or power infrastructure, resource coordination is especially necessary to prolong the network lifetime. This thesis investigates methods to reconfigure visual sensor networks to achieve a tradeoff between surveillance quality and resource consumption. First a formal description of the problem is presented. Based on this, the algorithmic design space of this problem is explored. A centralized approach using an evolutionary algorithm is presented for use in environments where no dynamic changes are expected. For environments with slow changes in the surveillance requirements, a distributed algorithm is described. For environments with fast changes, this algorithm is complemented with an object-handover algorithm. Besides algorithms, software tools which support reconfiguration in visual sensor networks are presented. This includes a software framework for sequential data processing and a distributed middleware system. In extensive evaluations the significant resource savings achievable with the presented algorithms are shown. Finally, the applicability of the developed evolutionary algorithm in a different field is shown. In a cloud server infrastructure, the load balancing is reconfigured to achieve cost savings without lowering the service quality.