Clustering based fuzzy logic for multimodal sensor networks: A preprocessing to decision fusion
Journal of Ambient Intelligence and Smart Environments
| Hi-index | 0.00 |
Wireless sensor networks have the integrated ability to observe phenomena, process data, and communicate information. This emerging technology has the potential for revolutionizing wide range of applications. Acoustic signals convey rich information about the acoustic source and the propagation media, and thus have been widely used for monitoring applications. Wireless sensor networks can enable many novel acoustic monitoring applications such as the non-intrusive, simultaneous monitoring of the interaction among multiple animals in the wild using animal vocalizations. The key to realizing the full potential of wireless sensor networks is collaborative signal and information processing (CSIP) that fuses data from multiple sensors and adapts sensor collaboration to the ever-changing targeted phenomenon and system state. The choice of CSIP algorithms has a profound impact on not only information processing quality such as estimation accuracy and robustness but also system operation quality such as system scalability and utilization. We used wireless sensor networks for acoustic target monitoring as a case study to investigate the design and implementation of the CSIP algorithms and the distributed computing with an emphasis on the information-directed collaborative target localization and tracking. We have investigated a broad range of CSIP issues in sensor networks. We have implemented and tested different acoustic beamforming methods for target localization both in free space and with limited reverberance using wireless networks of commercial-off-the-shelf (COTS) embedded computers. We have solved the microphone spacing problem for robust direction-of-arrival (DOA) estimation that uses the approximate-maximum-likelihood (AML) algorithm. We have designed a highly effective audio data reduction algorithm to reduce across-node wireless communication in real-time acoustic beamforming. We have invented a light-weight real-time sensor selection heuristic that is suitable to sensor network platforms with moderate computational power. We have discovered that the minimum entropy of the posterior target location distribution is am effective measure to quantify the lower bound of target localization uncertainty that could be achieved by sensor networks. Using the minimum entropy of the posterior target location distribution, we have analyzed the impact of the sensor network deployment geometry on localization accuracy, and derived a sensor placement strategy to achieve high localization accuracy in a given region to monitor. In addition, we have devised a staged event-driven model for collaborative information processing that filters out irrelevant events as soon as they are recognized as irrelevant.