Vision-based Indoor Lighting Assessment Approach for Daylight Harvesting

Electricity-based lighting is a major source of energy consumption in buildings, which in turn has great potential to reduce its energy consumption by incorporating lighting control strategies, and it is highly recommended to evaluate energy savings before implementing such strategies. Thus, this study focuses on the development of a vision-based modeling approach to assess indoor lighting (VILM) from daylight, dispensing a network of expensive photosensors. The proposed VILM uses self-supervised feature extraction and machine learning algorithms including support vector machine (SVM), logistic regression (LR), and Random Forest (RF) to learn each time-variant scene feature and establish the required illuminance model. Experimental tests were conducted in a university building, laboratories, and classrooms equipped with cameras. The results obtained showed that the RF model yielded the best accuracy and predicted the speed of indoor lighting level compared to other models. In sum, lighting evaluation before integrating control is a very important module for managing and achieving energy-efficient lighting.