Towards Window State Detection Using Image Processing

As industry pushes towards more sustainable design, an essential tool that helps designers quantify energy usage of a building is building energy simulation. However, researchers have noticed that there is a gap between the estimated and measured energy use. One aspect contributing to this discrepancy is inaccurate data-driven occupancy behaviour models. To create better models, researchers need to obtain data in a new method that does not produce inauthentic data. Ex-situ camera-based occupant behaviour monitoring has been proposed as a solution. This study uses image processing technologies to identify the windows on a façade and determine their individual state (i.e. open, partially open, or closed). The algorithm developed in this study yields a 90% accuracy rate over all the windows tested. This algorithm is specifically targeted at punched façades with awning windows. Factors that affected the accuracy of ex-situ camera-based occupant behaviour monitoring include environmental conditions such as lighting, obstructions, and reflections. Furthermore, there are challenges in thresholding and identifying the significant peaks for window angle image data. The next steps of this research should determine appropriate threshold values through additional testing and explore new image-processing techniques for other window types.