Random forest learning method to identify different objects using channel estimations from VLC link

This paper demonstrates the feasibility of using supervised learning algorithms to identify the presence of different objects, taking advantage of the effect that they create on the VLC channel gains. For this purpose, a software-defined VLC link is implemented using a Phosphor-converted LED, whose light intensity is modulated by an Optical OFDM frame that includes synchronization words and pilot sequences for channel estimation. Actual estimated channel gains are collected in the receiver, which are used to train and assess the performance of the Random Forest classifier. The accuracy of the monitoring system is evaluated using three different objects, showing an accuracy in the order of 90% in detecting the objects, even when they take different positions when obstructing the VLC link.