Fundamental challenges in designing environments with media-rich ambient services involves not only the development of appropriate sensing technologies, but as importantly, the implementation of a distributed media processing system which can process, integrate, and leverage the sensed data in real time to provide the various services. In recent years, a great deal of progress has been made in media service workflow processing systems. In most existing solutions, however, the workflow nodes, which operate on the data, are selected out of a centrally assigned candidate pool. These candidate organizations cause either extensive resource provisioning or poor-quality operator mapping between logical workflow nodes and the available physical resources nodes. Consequently, instantiating a media processing workflow to the underlying hardware before the workflow execution begins does not lends itself to adaptive and autonomous operation of the workflow, scalable to resources and demand.In this paper, we propose a novel decentralized multimedia workflow processing system, DANS, in which operators defined in workflows are mapped into (distributed) physical nodes through Distributed Hash Table (DHT)-based overlay substrate in a purely decentralized and adaptive manner. The redundancy in the system, in terms of availability of multiple nodes able to perform the same task, enables the system to scale with demand. Furthermore, physical workflow nodes (operator instances) are able to locate and select the next filter or fusion operator instance autonomously, while ensuring the correct execution of the workflow.