Video Content Analysis-Based Detection of Occupant Presence for Building Energy Modelling

The information on occupant presence plays a critical role in building energy modeling for spaces with a high number of occupants. A thorough understanding of occupant behavior is key to precise Building Energy Modeling (BEM) and to increase the precision of the simulation results. Capturing occupant-related information is difficult due to its stochastic and temporally uncertain nature. In this paper, we propose a robust video content analytical approach for the fast and accurate analysis of temporal and spatial video content. This approach counts the number of occupants in a classroom in an existing building by processing the recordings of video cameras. Two novel counting methods were implemented. The first, namely the Average Counting Method, uses cameras installed in the room directed in different angles, This method relies on detecting and counting occupant heads using a deep convolutional network, namely YOLOv2, that we trained on an existing head dataset. The second method, namely the Entrance Counting Method, uses cameras directed towards the room entrance and increments or decrements a counter based on the occupants entering and exiting the classroom. In addition to YOLOv2, the Discriminative Correlation Filter with Channel and Spatial Reliability (CSR-DCF) was used to create temporal relationships. At the same time, the ground truth was established by manual head counting. The analysis of the results of the one-week recordings initially indicate occlusion problems for the videos of door cameras in case of crowded groups. The videos of room cameras also experienced similar difficulties due to occlusions and the detection of occupants located further from the cameras. Based on these observations, an approach to combine the calculations of both methods is developed, wherein the room cameras are considered as the reference in case of local minima, while the rest is calculated using door cameras with respect to these references. Finally, we validate our approach through two experiments. The first experiment concerns the quantitative comparison between the proposed approach and the ground truth acquired through manual counting methods. The second experiment evaluates the results of the proposed approach in an energy model by quantifying the degree of change in terms of different metrics concerning building energy performance. The results are indicative of the critical role of occupancy in energy modeling.

Ipek Gürsel Dino, Esat Kalfaoğlu, Alp Eren Sarı, Şahin Akın, Orçun Koral İşeri, A. Aydın Alatan, Sinan Kalkan,  Bilge Erdoğan


Popular posts from this blog

07.09.2017 - How to create or modify EnergyPlus Weather Data (EPW)

An Algorithm for Efficient Urban Building Energy Modeling and Simulation