Data and code for "Characterisation of material optical properties for indoor daylight simulation"

This repository contains the data and code for the PhD thesis chapter "Characterisation of material optical properties for indoor daylight simulation."Summary: This chapter evaluates three image-based techniques for on-site material reflectance characterisation in daylight simulations—Illuminance-Proxy, Ferwerda’s phone-based method, and Luo’s learned SVBRDF estimation—and validates a spectral uplifting approach for non-visual simulations. A meeting room with diverse finishes was measured using a Konica Minolta CM-26dG spectrophotometer (SCI/SCE) to establish ground truth. The Illuminance-Proxy reconstructs wall reflectance maps from HDR luminance and sparse reference points (kriging); Ferwerda’s technique estimates tristimulus reflectance from unconstrained phone images; Luo’s technique infers diffuse/specular/roughness from flash-lit phone photos. Accuracy is reported using NMBE, NRMSE, ∆Eₐᵦ, and room-scale metrics (TAI, UDI, DGP). Ferwerda’s technique shows the most balanced performance (NMBE ≈ 13%, NRMSE ≈ 31%), stable across lighting conditions, with TAI overestimation of 33–65%. Illuminance-Proxy yields NMBE ≈ 19%, NRMSE ≈ 41%, but overpredicts TAI by ∼101% in the case study; a 3-channel HDR test indicates per-point ∆ρ < 0.08. Luo’s technique exhibits the largest errors (NMBE ≈ 25%, NRMSE ≈ 62%), with ∆Eₐᵦ up to ∼60 and TAI overestimation up to ∼134%. All techniques predict low DGP with negligible deviation (NRMSE < 0.02). Spectral uplifting reconstructs smooth, energy-conserving spectra with wavelength-wise errors typically < 20%, yielding melanopic equivalent illuminance (EML) differences < 6%. Overall, Ferwerda’s technique offers the best accuracy–practicability trade-off, Illuminance-Proxy is useful for planar diffuse surfaces when sparse ground truth is available, and spectral uplifting enables non-visual analyses from tristimulus inputs with minimal error.