Buildings are a corner stone of our society: we spend most of our time in buildings, and the building sector accounts for most of the energy demand and CO2 footprint. Improving indoor comfort and energy efficiency of the built environment is thus one of the main key to tackle our current environmental and social challenges. Because buildings affect so many aspects of our life, multiple disciplines, both technical sciences and human sciences, study these objects.
Poor indoor environment quality is, unfortunately, fairly common in most of our buildings. This takes a tremendous toll on the comfort, well-being, health and productivity of most of us, inducing massive extra costs for our social and health systems and massive losses for the companies and our economies. Among these, glare (cause by bright beam of sunlight striking the eyes of building occupants) is a common source of visual discomfort. Office workspaces are especially prone to glare discomfort impairing occupants’ satisfaction and productivity.
AI approach to study of glare
Glare is very hard to assess and predict with fixed light sensors because it is affected by multiple factors such as position and orientation of the occupant in the room, position and orientation of the windows, position of the sun in the sky, furniture layout, surfaces reflecting or blocking light rays and cloud cover. Moreover, glare discomfort can appear and disappear within few minutes or seconds in countries with rapid cloud movements such as Denmark.
In this project, a very different approach is taken to assess whether or not a person is experiencing glare discomfort. Instead of relying on light sensors, the face of the occupant is directly used as a subjective probe to assess if visual discomfort is occurring. Indeed, humans react in a very distinctive way when experiencing glare: squint eyes, clear facial expression. These facial expression features can be detected by AI performing computer vision. Indeed, machine learning and computer vision have matured considerably over the last years and accurate facial expression analysis is now possible. State-of-the-art machine learning algorithms are now available for various engineering applications with minimum training effort as most of the basic functionalities are already pre-trained, well understood and well documented.
However, to develop an AI algorithm for detecting glare discomfort in the built environment, it is necessary to assemble a multidisciplinary R&D team with good skill in visual comfort, machine learning methods, programming, data analysis, experimental investigations, control and automation. Fortunately, such interdisciplinary R&D group was rapidly set through the good intercommunication between AAU-BUILD and AAU-CREATE research teams. AI for people bridging project fund has rapidly validated the idea and a clear project plan has been established with the objective of developing a proof-of-concept prototype showing that such a method to estimate subjective glare discomfort is possible.
AAU-BUILD created the experimental setup to generate and acquire the data necessary to train, test and validate the AI algorithm. AAU-CREATE then selected and trained the AI algorithm with different machine learning methods. Finally, AAU-BUILD integrated the AI algorithm into a control interface that serves as a demonstrator and provides feedback to regulate a shading device. A working prototype has thus been created in 3 months. The AI algorithm has an accuracy of 90%, which is comparable to other similar computer vision applications.
With the help of a daylight and glare specialist from DTU, this project was presented at the conference article for CISBAT 2021.. This innovative approach is very promising to tackle many problems encountered by research teams working on glare, which resulted in the authors of this study receiving the “best paper award” of this CISBAT 2021 conference.[AS1]
This is a major achievement and a great encouragement to continue exploring and developing AI-based solution for improving the sustainability of the building sector, and reinforce interdisciplinary collaborations at Aalborg University.