Indoor Visible Light Positioning Based on Extreme Learning Machine with Local Training Samples

Recently, the visible light positioning (VLP) has become a promising indoor positioning method due to its high positioning accuracy and low cost. However, the computational complexity and the positioning error of VLP based on machine-learning (ML) algorithms should be seriously considered, especially when a large number of training samples are used. In this paper, we present a VLP method based on the extreme learning machine (ELM) with local training samples (ELM-LTS) which are distributed around the target node (TN). First, the region iterative threshold reduction method is proposed to obtain the square region where the TN is located based on the large training samples. Then, the local training samples in the square region are selected to train the ELM network. Finally, the position of the TN is estimated accurately by using the trained ELM network and its received signal strength. Simulation results demonstrate that the averaged positioning error (APE) of the ELM-LTS is lower than that of the ELM using large training samples.