Occupancy Estimation in Enclosed Spaces using an Indirect Approach, laying the Foundations to Build an IoT Architecture

Abstract

The buildings industry accounts for 30% to 40% of total consumed energy worldwide, and with most of this energy coming from fossil fuels, improving energy efficiency is critical to reducing the harmful effects of this industry on the environment. Fortunately, opportune information about the number of occupants has been identified as a significant contributor to improving energy efficiency. The several works that have been carried out to solve the problem of occupancy detection/estimation fall in one of the following categories: (1) direct approaches based on sensors and cameras to measure occupancy directly, and (2) indirect approaches based on environmental data to derive the occupancy information. Due to the cost and privacy issues, indirect approaches are preferred for most use cases. This thesis focused on estimating occupancy in buildings’ indoor spaces using environmental variables andMachine Learning techniques. Specifically, the use of temperature, humidity, and pressure information was proposed to estimate the level of occupancy. Additionally, feature selection and time resolution selection steps were used to achieve high accuracy. In the process, it was necessary to generate a dataset with occupancy information from two different locations with contrasting characteristics. This dataset is an essential contribution as no other dataset suitable for estimating occupancy using the proposed environmental variables is publicly accessible.Likewise, a review of IoT platforms was carried out to identify the components required to build an occupancy estimation system. Among the contributions, it is reported that at least98% of accuracy can be achieved using this approach and a kNN model. Also, a theoretical architecture for an occupancy estimation system using AWS IoT Core was documented. Finally, the generated dataset was made publicly accessible through the Mendeley Data repository.