Previous Abstract Return to Session C6 Next Abstract

Session C6: Terrestrial Signals of Opportunity-Based Navigation Systems

Indoor Localization Based on Machine Learning-Assisted PDR and Signals of Opportunity From Ambient Generic BLE Devices
Masakatsu Kourogi, Akihiro Sato, Satoki Ogiso, Ryosuke Ichikari, and Takashi Okuma, AIST

Indoor positioning is crucial for smartphone users in areas where GNSS signals are unavailable, allowing them to be guided and navigated through complex environments like large shopping malls, airports, and train stations. While Bluetooth Low Energy (BLE) beacons offer potential for indoor positioning, deploying and managing them on a large scale can be challenging and labor-intensive. This paper introduces an innovative method that utilizes Pedestrian Dead-Reckoning (PDR) along with ambient BLE devices, including lighting fixtures, digital speakers, and signage, to locate smartphones. The proposed method employs a build-and-refine approach to create a map of the BLE devices whose locations are initially unknown. In the initial stage, constraints from the floor map’s skeleton structures and the walking speed are used to estimate the absolute trajectory from the PDR-based relative trajectory, which is then used to generate the first-generation BLE maps. After obtaining the BLE maps, PDR-based trajectories are combined with BLE scan data to estimate absolute trajectories, which are then used to create the next-generation BLE maps. To estimate walking speed, we use machine learning (ML) approaches to fuse multiple features such as power and frequency of acceleration patterns in the frequency domain, and time-integrated acceleration in the time domain. Evaluation conducted on the train stations and shopping malls where GNSS localization is completely unavailable. We found that the estimation of walking speed is dramatically improved with ML-based regression compared to simple linear regression, and the median localization error of the smartphones is below 2 meters.

Previous Abstract Return to Session C6 Next Abstract