Accuracy Evaluation of Indoor AoA Visible Light Positioning With MLP Regression Model

Visible light is considered as a promising indoor positioning resource due to its ubiquitous coverage, safety, and license-free. As most of the indoor visible light positioning methods focus on the received signal strength and adopt classification models, which makes their positioning accuracy severely limited by the grid density of measured points. A positioning method based on the angle of arrival (AoA) and using a multi-layer perceptron (MLP) regression model is proposed in this letter. In the offline stage, the data set is constructed by the measured AoA of the incident light signal with respect to different light sources, then preprocessed and fed into the MLP model to obtain the optimal hyperparameters. In the online stage, immediately measured AoA are fed into the optimized MLP model, and the receiver's position can be accurately estimated in real time. An indoor positioning platform with 4 light sources and an OpenMV camera is built, and the experimental results show that the proposed positioning method could achieve a 5.2 cm positioning error for 80% measured points, and has a robust generalization capability.