Classification of artificial light sources and estimation of Color Rendering Index using RGB sensors, K Nearest Neighbor and Radial Basis Function

Abstract

Abstract Three types of artificial light sources work with electricity: incandescent, fluorescent and LED. These sources require characterization processes to allow selecting the most suitable for the application, to evaluate their capacity or more recently to tune and adjust their replicability using control algorithms. Therefore, it has been necessary to develop indices that represent these capabilities. The Color Rendering Index (CRI) is a measure used to characterize the color reproducibility of a light source in comparison to an ideal light source. The Correlated Color Temperature (CCT) is used to characterize light sources by representing the color as the temperature of a black body in Kelvin that shows nearly the same chromaticity as the analyzed light source. Using spectral information to determine the values in the XYZ space and deriving the calculation described in the standard is the best way to estimate the value of the CCT and the CRI. In this work, we implement a method to classify light sources and to select an estimation model of the CRI and the CCT using a low cost RGB sensor. The model estimation has been developed in this work and a separated algorithm for each source type has been built. The results show that using a K-Nearest Neighbor classifier, the error resulted less than $3.6%$. The error of the model estimation for the LED was 1.8%, for fluorescent light sources 0.09% and 1.2% for incandescent light sources.