Output details
11 - Computer Science and Informatics
University of St Andrews
Using situation lattices in sensor analysis
<02>Presented at the leading IEEE conference in the area (<20% acceptance) this paper improves on traditional sensor fusion algorithms by using a structured mathematical model of the real-world situation being classified. Such situation recognition is the critical scientific problem in pervasive computing and sensor data analysis. Comparative evaluation against other approaches over a widely-used smart-home dataset shows significantly improved recognition accuracy. The technique also provides a further major improvement over other algorithms by highlighting where recognition fails, allowing designers to deploy additional sensors to discriminate between otherwise indistinguishable situations. The lattice technique forms the basis for several modern systems.