Particle Filter-Based Enhanced Transition Model in Signal for Unsupervised Localization

• Published in: IEEE sensors journal Vol. 24; no. 21; pp. 35845 - 35857
• Main Authors: Chen, Hailong, Shen, Xingfa, Wang, Yongcai, Shi, Nan
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.11.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Algorithms; Background noise; Clustering; Computational modeling; Data augmentation; Data models; Errors; Fingerprint recognition; Indoor localization; Inertial coordinates; inertial navigation; Inertial platforms; Kalman filters; Linear systems; Localization; Location awareness; Noise pollution; Nonlinear systems; particle filter; Phase matching; Pollution sources; Random noise; Sequences; Signal strength; Synchronism; Time measurement; Trajectory; transition model (TM); Wi-Fi; Wireless fidelity
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2024.3439540

Unsupervised indoor localization methods have garnered significant attention for their low training costs and minimal active participation requirements. A crucial approach in this field is positioning by synchronization of Wi-Fi signal strength and inertial measurement [received signal strength (RSS) + inertial measurement unit (IMU)] sequences using mobile devices, which relies on abundant, disparate, and unlabeled RSS + IMU sequences to establish continuous time interval mappings between two signal state spaces for unsupervised positioning. Errors and noise pollution in sensor measurements are the main sources of inaccuracies in unsupervised indoor positioning. Therefore, the development of suitable algorithms to mitigate errors caused by environmental factors is a key research focus. The transitional model in signal space (TMS) represents a notable method based on this model. KF-TMS algorithm reduces Gaussian noise interference on measurements and positioning accuracy when using Kalman filter (KF) in linear system environments. This article introduces the Enhanceded Particle Filter TMS (EPF-TMS) algorithm, which has been designed for nonlinear systems and non-Gaussian environments. The proposed approach addresses the limitations of KF-TMS in terms of practical application scope and accuracy. The EPF-TMS algorithm enhances the mapping between continuous RSS signals and single-step motions through trajectory data augmentation and filter direction matching in the offline phase of EPF-TMS, which improves the robustness and localization accuracy of the model. In addition, this algorithm addresses the limitations of traditional transitional model-based methods through data clustering processes and data dimensionality reduction processing, which are constrained by extensive computational demands and high time complexity. Comparative experiments have demonstrated that the EPF-TMS algorithm has superior precision in both simulated and real-world experiments, effectively minimizing the impact of environmental noise on positioning. Moreover, the EPF-TMS algorithm demonstrates a notable advantage in terms of time efficiency following the completion of the offline processing phase, thereby reducing the overall time required for the localization process.