Publication Year: 2018 Publication Type : JournalArticle

Abstract:

Research on building and room occupancy counting is becoming more important. By understanding and knowing the numbers of people within a building, the heating, cooling, lighting control, building energy consumption, emergency evacuation, security monitoring and room utilisation can all be made more efficient. Thorough research in this area has implemented with various methods including the use of ambient sensors. However, occupancy models that have been studied in previous work require the use of many sensors, which is expensive for installation and on-going operation. In this experiment, we use a single sensor that is commonly available in the Australian BMSs to reduce the cost and complexity as more sensors mean less reliability. The research of using a single type of information such as CO2 and inferring it to predict human occupancy is novel. Hence, many possibilities can be explored by using this technique. Furthermore, our model can be trained to have a high accuracy with training data gathered over only two weeks. This algorithm is robust in handling both low and high occupancy number up to three hundred occupants. Furthermore, with CO2 data, the privacy of every single individual is protected as no personal information is required. This method is device-free in a notion that no device would be attached to the body throughout the experiment phases. Our method produces on average 8.46% better accuracy compared to the baseline method. In addition, our prediction model reduces the accuracy by 6.9% after predicting more than 10 days.

BibTex:

Cite: