Human behavior analysis frameworks using IoT sensors for data-driven workplace design

Abstract

Recently, workplace design has aimed to efficiently use space to promote environmental sustainability and reduce operating and rental costs. It has also aimed to improve occupant satisfaction and increase face-to-face (FTF) interactions because both of these elements are closely correlated with organizational outcomes. However, in many cases, workplace design fails to achieve its aims because it depends on conventional design guidelines or an architect’s subjective knowledge and intuition without being based on objective information regarding how users actually use the space. In order to address this limitation, data-driven workplaces design based on accurate recognition of user activity is required because it provides objective information related to users’ space use which helps architects make better design decisions. In recent times, advanced sensor technologies have a great potential for use in activity recognition because they can automatically measure objective data associated with user activities. Despite the potential of these technologies, there is no existing method to accurately recognize user activity for data-driven workplaces design. Thus, in this thesis, I developed two frameworks that recognize user activity based on objective sensor data. Specifically, a turn-taking pattern based interaction estimation (TIE) framework that recognize FTF interaction of users, and a framework to recognize user specific activity pattern generation (UAPG) were developed. In the demonstrations, TIE and UAPG frameworks achieved 77.7 and 91.8 percent of activity recognition accuracy respectively, and provided objective information associated with users’ space use for data-driven workplace design.