MMGraphSLAM: Autonomous indoor positioning based on millimeter Wave GraphSLAM

Currently, it is still challenging to design an indoor positioning system with high precision and ubiquitousness. Recently, more attentions have been paid to the application of millimeter wave (mmWave) Radar because of its high sensing accuracy, autonomy and ubiquitousness. This work presents an autonomous indoor pedestrian positioning system based on mmWave GraphSLAM (MMGraphSLAM) without additional beacon infrastructure. In MMGraphSLAM, trajectories based on inertial sensor are optimized by mmWave Radar and GraphSLAM. We use PointNet Auto-Encoder neural network to enhance the point cloud data from mmWave Radar. Moreover, MMGraphSLAM fuses the point cloud and intensity-range information to improve the accuracy of loop closure detection. Then feature information is allocated with different weights to guarantee the reliability of back-end in MMGraphSLAM. A set of comprehensive evaluations in multiple experiments illustrate that our proposed system can better recover pedestrian’s walking trajectories and reach a sub-meter positioning accuracy without additional signal infrastructure.