Distributed Fusion of Sensor Data in a Constrained Wireless Network

Smart buildings with connected lighting and sensors are likely to become one of the first large-scale applications of the Internet of Things (IoT). However, as the number of interconnected IoT devices is expected to rise exponentially, the amount of collected data will be enormous but highly redundant. Devices will be required to pre-process data locally or at least in their vicinity. Thus, local data fusion, subject to constraint communications will become necessary. In that sense, distributed architectures will become increasingly unavoidable. Anticipating this trend, this paper addresses the problem of presence detection in a building as a distributed sensing of a hidden Markov model (DS-HMM) with limitations on the communication. The key idea in our work is the use of a posteriori probabilities or likelihood ratios (LR) as an appropriate "interface" between heterogeneous sensors with different error profiles. We propose an efficient transmission policy, jointly with a fusion algorithm, to merge data from various HMMs running separately on all sensor nodes but with all the models observing the same Markovian process. To test the feasibility of our DS-HMM concept, a simple proof-of-concept prototype was used in a typical office environment. The experimental results show full functionality and validate the benefits. Our proposed scheme achieved high accuracy while reducing the communication requirements. The concept of DS-HMM and a posteriori probabilities as an interface is suitable for many other applications for distributed information fusion in wireless sensor networks.