Spatiotemporal occupancy in building settings

Abstract

This thesis presents an investigation of methods to capture and analyze spatiotemporal occupancy patterns of high resolution, demonstrating their value by measuring behavioral outcomes over time. Obtaining fine-grain occupancy patterns is particularly useful since it gives researchers an ability to study such patterns not just with respect to the geometry of the space in which they occur, but also to study how they change dynamically in time, in response to the behavior itself. This research has three parts: The first is a review of the traditional methods of behavioral mapping utilized in architecture research, as well as the existing indoor positioning systems, offering an assessment of their comparative potential, and a selection for the current scenario. The second is an implementation of scene analysis analyses using computer vision to capture occupancy patterns on one week of surveillance videos over twelve corridors in a hospital in Chile. The data outcome is occupancy in a set of hospital corridors at a resolution of one square foot per second. Due to the practical detection errors, a two-part statistical model was developed to compute the accuracy on recognition and precision of location, given certain scenario conditions. These error rates models can be then used to predict estimates of patterns of occupancy in an actual scenario. The third is a proof-of-concept study of the usefulness of a new spatiotemporal metric called the Isovist-minute, which describes the actual occupancy of an Isovist, over a specified period of time. Occupancy data obtained using scene-analyses, updated with error-rate models of the previous study, are used to compute Isovist-minute values per square feet. The Isovist-minute is shown to capture significant differences in the patient surveillance outcome in the same spatial layout, but different organizational schedule and program.