Chameleon: Adaptive Sensor Intelligence for Smart Buildings

Abstract

Diversity in building types and evolving usage patterns within homes, offices, and schools can be a big hurdle for smart city infrastructure. Smart sensors are stifled by long calibration procedures, fixed preprogrammed decision making, and the need for large labeled data sets. These problems introduce tradeoffs between system accuracies, scalability, and adaptability for the Smart Building IoT infrastructure. This article presents Chameleon, an adaptive sensor fusion and hybrid machine learning architecture that is able to classify room activity states. We test the performance in a classroom environment and an office environment with completely distinct geographical and layout characteristics. We evaluate the system with a total of eight weeks of experiment data and a total of 16 training and clustering sequences. Experiments show that the system is able to correctly classify activity states with accuracies in the range of 87%–99% on test sets. We show that the system is able to keep its high accuracies in a way that makes it adaptable to different rooms and scalable due to minimal hardware and software requirements.