LightSplat: Leveraging Gaussian Splatting for Illuminance Estimation in Light Audits

Lighting significantly impacts building energy consumption and is crucial for worker productivity and well-being. However, traditional lighting audits are time-consuming and often limited by sparse, inconsistent measurements from light meters. We introduce LightSplat, a novel approach for estimating indoor illuminance using only RGB images. LightSplat first trains a Gaussian splatting (GS) model to reconstruct a 3D representation of the indoor space, allowing for image rendering from arbitrary viewpoints. A Monte Carlo-based illuminance estimation module then computes a dense illuminance map across the 3D model. We evaluated LightSplat in a real-world institutional office under two lighting conditions, using light meter data as ground truth. The method achieved mean relative errors of 23.28% and 19.15%, demonstrating strong accuracy. LightSplat addresses key limitations of conventional lighting audits by providing dense, reliable lighting data efficiently. LightSplat shows promise as a valuable tool for lighting design, modeling, and building energy optimization.