Reliability-driven sensor fusion with application in automatic signature verifications

The reliability of Intelligent Sensory Systems with multi-failure modes is a research area that at the present time lacks adequate attention. In this thesis a novel formulation for assessing and optimizing the reliability of multi sensory systems is proposed. This formulation is capable of modelling a system with m failure modes, under three operational schemes. For each operational scheme an analytical reliability assessment model is developed. Adapting the Voting as a fusion technique, heuristic techniques and genetic algorithms are developed to optimize these models under the different modes of operation. Waveform classification and signature verification problems are used to explain and demonstrate the potential of the proposed formulation.; Experimental results showed that without optimization the performance of multi-sensor multi-failure mode systems slightly underperformed the performance of some individual sensors. However, the performance of the heuristically optimized versions of these systems outperformed the performance of the best sensor used in such systems. This was true for both waveform classification and signature verification problems. Furthermore, by optimizing signature verification system using genetic algorithms a state-of-the-art performance of 96% was reached.