A Multi-Sensor Fusion Algorithm for Pedestrian Navigation Using Factor Graphs

Real-time and robust navigation for pedestrians is a challenging problem in the satellite denial environment. When designing an autonomous navigation framework for pedestrians, one often estimates local pose with either the visual odometer or foot-mounted MIMU. However, the framework has poor robustness in diverse environments owing to the shortcomings of vision and inertia. In this paper, we propose a pedestrian navigation algorithm to estimate local pose by fusing foot-mounted MIMU with cameras and optimize the state with other sensors. Different from sensors rigid connection, the local sensors are attached to the pedestrian, which is a flexible body but has a similar position estimation in diverse parts of the body. The global state is optimized in batch with the factor graph. Therefore, the accumulated drifts of inertial integral are corrected and the robustness of system is improved. We evaluate the performance of our algorithm with real-world experiments. Results are compared against other algorithms to show that the absolute trajectory accuracy in the algorithm has been greatly improved.