Magnetic Field-Based Localization in Factories Using Neural Network With Robotic Sampling
Ting-Hui Chiang, Zao-Hung Sun, Huan‐Ruei Shiu, Kate Ching‐Ju Lin, Yu‐Chee Tseng
Abstract
With the advent of Industry 4.0, localization of materials and factory items will play important roles in factory automation. Since GPS signals are not available in indoor environments, a lot of indoor localization technologies have been proposed based on inertial sensors, audio signals, visible light, wireless signals, etc. In this research, we consider using magnetic fields, which usually exhibit high uniqueness at different locations especially in factories where a lot of stacks, machineries, materials, and metal partitions may coexist. These factors allow us to incorporate deep learning neural networks to learn location-related features. Existing works try to collect magnetic field data by human and leverage interpolation to augment dataset. However, our experiments show that the data generated by interpolation is usually different from the ground truth because magnetic fields may not be linear. Therefore, to collect a large enough dataset without human intervention, we dispatch a robot carrying a smartphone to collect dataset at a fine resolution. We use these collected data to train two localization models: deep neural network (DNN) and recurrent neural network (RNN). Besides, we augment our RNN training dataset by combining multiple single-point magnetic values to synthesize fake magnetic trajectories. We conduct field trials, which validate that our approach outperforms previous work.