Robustness enhancement of indoor visible light positioning system based on cascaded residual neural network

Abstract

We proposed and experimentally demonstrated an ensemble learning scheme based on cascaded residual neural network (CRNN) to enhance the robustness of machine learning (ML) based visible light positioning (VLP) systems under model-shake. The training complexity of the CRNN is low. Experimental results reveal that with an acceptable complexity cost, the positioning accuracy under model-shake can be optimized from 8.31 cm to 2.04 cm and 1.56 cm with the proposed 2nd CRNN and 3rd CRNN, respectively.