Fusion of IMU and Probabilistic Model for Indoor Localization Based on Bayesian Framework

High-accuracy indoor localization is a key enabler of ubiquitous location-based services (LBSs) in the Internet of Things (IoT), with applications in mobile robots, asset tracking, and beyond. For indoor localization, it has been reported that the methods based on probabilistic models have high localization accuracy and strong generalization in the presence of nonline-of-sight (NLOS) conditions and multipath effects. To further leverage such advantages, this article proposes two fusion localization methods based on Bayesian filters which fuse an inertial measurement unit (IMU) motion model with a probabilistic model constructed by soft information (SI) framework to enhance localization performance. First, we propose a method based on particle filter (PF) to directly fit the posterior probability density distribution (PDF), called PF-SI. This method reduces accuracy loss caused by linearization and achieves high accuracy. Then, to reduce the high computational complexity of the PF-SI method, we utilize an error state Kalman filter (ESKF) to construct linearized error state transition and error observation equations and update the filter with distance residual, as ESKF-SI. This method has slightly lower localization accuracy but significantly improves computational efficiency. Finally, experimental results in a real indoor scenario based on ultrawideband (UWB) signals are presented. The results show that the two proposed fusion methods can achieve a root mean square localization error of less than 0.25 m in a complex NLOS scenario.