Visual analytics for large scale ethnography: Exploring computational and ethnographic analysis techniques to better understand the occupation and utilization of the built environment

Observational data within Post Occupancy Evaluations (POE) has had the problem of being deemed as 'unreliable' due to the fact that it is not based on quantifiable, empirical evidence. Even though ethnographic observation is quite effective and has a method, the common misconception is that it is just someone watching and drawing conclusions. What I propose is a system that can turn some of the ethnographer's observational data into empirical data that can be better incorporated into traditional POE reports. The data becomes more credible in the eyes of those unfamiliar with ethnographic observational practices, and can be used to benefit current and future building designs.;Such a system utilizes principles of Computer Vision and Machine Learning to develop a means for collecting and synthesizing observational data. The system will use a camera array installed in a space to track the movement of the people within. Additionally, we will enlist the help of ethnographers to teach the program to look for 'meaningful interactions' of the people within the space. Based on these notable events, the program will be able to draw data from like events, store the data, and use it when examining future data. This system will allow for data collection to occur continuously within a building where such a system is installed, providing a continuous stream of information to be used in Evidence Based Design. In this way, the evaluation team will be able to present a new dimension of data in their POE for the architects and owners to consider.;The following thesis addresses background information on the structure of POEs, Ethnographic methods of observation, and proposes a preliminary User Interface for inputting information as well as a method for identifying 'meaningful interactions'.