Computational frameworks for context-aware hybrid sensor fusion

This paper proposes inexpensive, specialised, computational frameworks that automate and integrate context-aware sensing, data aggregation, information extraction and understanding and qualitative decision making through intelligent algorithms. Its contributions are spread across context-aware data collection and aggregation, hybrid feature extraction incorporating both supervised and unsupervised approaches, and decision-based information fusion. It provides a toolkit that makes it easier for applications to use context. It presents a hybrid feature extraction framework based on two diverse optimisation problems in aspects of risk and independence to extract features resulting in higher classification performance. It combines a context-aware multi-sensor data collection model and a "Feature Input Feature Output (FeI-FeO)" based fusion model with an intelligent classifier to create a "Feature Input Decision Output (FeI-DeO)" based pattern recognition system, which can classify targets by eliminating redundant contexts. The proposed frameworks achieve context-sensitive information fusion with higher accuracy, less energy consumption and greater fault tolerance in resource-constrained environments with data collected from distributed sensors.