Unsupervised Localization by Learning Transition Model

• DOI: 10.1145/3328936
• Publication Date: 2019-06-24T13:45:01Z
• Authors: Xuehan Ye; Shuo Huang; Yongcai Wang; Wenping Chen; Deying Li
• Keywords: SIGNAL (programming language); Tracking (education)

Nowadays, it becomes very convenient to collect synchronized WiFi received signal strength and inertial measurement (RSS+IMU) sequences by mobile devices, which enables the promising solution conduct unsupervised indoor localization without pain of radio-map calibration. To relax needs floor-map information or trajectory knowledge, this paper proposes learn a transitional model (TM), segments massive unlabeled train that captures expected relationship between {zt--1, zt } ut--1, where zt--1, are two consecutive states at t -- 1, ut--1 is one step motion calculated from data. We present both in space (TMS) predict change (TMM) represent different ways. In particular, sequences, smoothed nearest neighbours, so transition learns relative state triggered motion. Its distinctive features (1) no external knowledge needed; (2) can be continuously on-line refined as incrementally collected. KALMAN filter based user location tracking methods given for models. Experiments show method provides comparable accuracy with manually fingerprint calibration methods.