Smartphone Indoor Fusion Localization with Trust Region-Based Magnetic Matching

Yu Zhu, Yan Jia, Kaiyi Zou, Xiaoguang Niu

Abstract: Magnetic indoor localization has attracted great attention in recent researches because of its advantage of not requiring additional equipment. However, existing magnetic matching methods have limited stability and accuracy due to insufficient consideration of gradient non-convergence and local optimal solution problem during the gradient descent iteration process. Moreover, in areas with inadequate magnetic features, the magnetic localization can be unreliable, leading to significant errors in certain regions. Therefore, we design a magnetic matching method using the trust region to adjust the utilization of gradient information during the matching process, dynamically balancing convergence efficiency and accuracy. By the way, we propose a fusion localization method FTRM, which enhances the robustness and accuracy of localization by determining the fusion weights of magnetic matching and pedestrian dead reckoning in the localization system through uncertainty, leading to more stable and accurate localization. The experiments show that the fusion localization system attain average accuracy of 0.545 m in real-world scenarios, achieve a 17.3% improvement.