Conference Paper

Proceedings of Building Simulation 2019: 16th Conference of IBPSA

Predicting Annual Equirectangular Panoramic Luminance Maps Using Deep Neural Networks

Yue Liu ¹, Alex Colburn ², Mehlika Inanici ^1
1 University of Washington, United States of America
² Zillow 3D Research Group, United States of America

DOI: https://doi.org/10.26868/25222708.2019.210369
Abstract: Annual luminance maps provide meaningful evaluations for occupants’ visual comfort, preferences, and perception. However, acquiring luminance maps require labor-intensive and time-consuming simulations or impracticable long-term field measurements. This paper presents a novel method to accelerate annual luminancebased evaluations utilizing a deep neural network (DNN). From a small subset (5%) of high dynamic range (HDR) imagery, our method can predict annual panoramic luminance maps (with 360-degrees horizontal and 180degrees vertical field of view) within an hour. Unlike the fixed camera viewpoint of perspective or fisheye projections that are commonly used in daylighting evaluations, panoramas allow full degree-of-freedom in camera roll, pitch, and yaw, thus providing a robust source of information for an occupant’s visual experience in a given environment. The DNN predicted high-quality panoramas are validated against Radiance RPICT renderings using a series of quantitative and qualitative metrics. With the developed workflow, practitioners and researchers can incorporate long-term luminance-based metrics over multiple view directions into the design and research process without the lengthy computing processes.
Keywords: Daylighting, Annual Luminance Maps, Deep Convolutional Neural Networks, High Dynamic Range Imagery, Adaptive Visual Comfort
Pages: 996 - 1003
Paper: