Sensor Arrangement for Classification of Life Activities with Pyroelectric Sensors - Arrangement to Save Sensors and to Quasi-Maximize Classification Precision

This paper deals with the sensor arrangement for activity classification systems with a group of pyroelectric sensors to realize a system with high classification performance using as few sensors as possible. It targets the people living alone. We convert this discrete optimization problem, which means whether or not the system select a sensor position candidate, to continuous, convex, and sparse optimization problem, and solve it efficiently by extended multi-class LPBoost via column generation. For some examinations, we showed the advantage of this algorithm. We also confirmed the significance of automatic arrangement system by comparing the arrangement obtained by this algorithm with the arrangement obtained by human judge.