A New Map/Odometer Tunnel Positioning Method Using Random Forest-Derived Heading Estimation

Deng, Chenxi

doi:10.1109/ojits.2026.3681900

A New Map/Odometer Tunnel Positioning Method Using Random Forest-Derived Heading Estimation

2026-01-01

0

PRE

AI

DOI：10.1109/ojits.2026.3681900

Original

Original request for help

Save

Abstract

En 中文

In global navigation satellite system (GNSS)-denied scenarios, odometer/inertial navigation system (INS) integration for train positioning tends to diverge due to accumulated attitude errors caused by uncorrected sensor drift. To provide a heading without sensor drift and assist odometer dead reckoning, a map-based heading/odometer positioning method and its optimism are proposed. In the map-based heading/odometer positioning framework, the heading is obtained by projecting and searching on the digital track map and then computing it using the corresponding interval. With the heading and odometer velocity, the position is computed from the train derived distance. Theoretically, shorter intervals provide more accurate headings and positions. However, due to onboard storage limitations, the stored interval information must be reduced, which means longer interval lengths are more acceptable to fulfill the industrial requirement. Thus, in the optimism for this method, a random forest (RF)-based heading estimation strategy is introduced to obtain accurate headings from long-interval maps. To be detailed, it utilizes both long- and short-interval digital track maps during offline training. By redividing the long-interval map to establish a correspondence with the short-interval map that provides accurate headings, a mapping from long-interval map data to accurate headings can be constructed, and the RF model is used to learn this mapping. During online prediction, the RF model can estimate headings from the onboard long-interval map for positioning. Experiments and dataset collection were conducted on the Qinghai-Tibet Railway between Sanglie and Dangxiong stations. A digital track map was generated from real-time kinematic (RTK) positioning-derived trajectory data and resampled at 3 m and 100 m intervals to form short-interval and long-interval map databases, respectively. Experimental results show that the RF-based strategy achieves heading accuracy comparable to that obtained using short-interval maps and improves map-based heading/odometer positioning performance in GNSS-denied environments when only long-interval maps are available.

Keywords:

Antennas

Antennas and propagation

Radio frequency

Location awareness

IP networks

Receivers

Communication systems

Computer networks

Telecommunications

Mobile communication

Train positioning

digital track map

heading

random forest

A New Map/Odometer Tunnel Positioning Method Using Random Forest-Derived Heading Estimation

Abstract

Journal

Organization

Cited Papers