Machine Learning Based Illuminance Estimation from RGB Sensor in a Wireless Network

Illuminance measurement is a salient feature to evaluate the quality of lighting and its precise measurement using appropriate sensor is of utmost importance for any real-time scenario. Here we present, the estimation of lux value on the acquired dataset using different types of machine learning regression models viz. Multiple Linear Regression, Support Vector Machine Regression, General Regression Neural Network, and Gaussian Process Regression. We have carried out extensive comparative performance assessment of the evaluating parameters: Percentage Absolute Error, R-Squared Error, Mean Absolute Error, Mean Absolute Percentage Error, and Root Mean Squared Error to predict the illuminance values with respect to a calibrated meter. The RGB values are obtained from a sensor that is integrated with a microcontroller and the lux data are obtained from a standard chroma meter at the client end. Now, the RGB data are transmitted in a wireless network and four different types of machine learning regression techniques are applied at the server end for proper estimation of the illuminance values. Thus, an accurate RGB sensor model is developed that can predict the lux values and also comparable to a standard lux meter. Gaussian Process Regression has given the best possible response for prediction of illuminance values in comparison with other regression methods on a real-time captured dataset.