An analysis of different sensors for turnout detection for train-borne localization systems

Abstract

Safe railway operation requires a reliable localization of trains in the railway network. Hence, this paper aims to improve the accuracy and reliability of train-borne localization systems proposed recently. Most of these approaches are based on a global navigation satellite system (GNSS) and odometers. However, these systems turned out to have severe shortcomings concerning accuracy and availability. We believe that the ability to detect turnouts and the branching direction thereon is the most valuable clue for improvement. Knowing the branching direction provides topological information about the train position. Thus, it complements the geographical information of GNSS and the longitudinal position information of odometers in an ideal way. With such a sensor setup a track-selective localization would be possible even if GNSS is unavailable or disturbed. Therefore, this paper compares the individual benefits of different sensor principles for turnout detection such as inertial measurement units (IMUs), cameras, and lidar (light detection and ranging) sensors. As a consequence, we focus on lidar sensors. For those we define requirements, review the market, and report the results of a case study in a tramway scenario. We proved that it is possible to detect rails, turnouts, and platforms. Finally we discuss our findings intensively and give an outlook on our further research.

Bayesian Train Localization with Particle Filter, Loosely Coupled GNSS, IMU, and a Track Map

“…The eddy current sensor, in principle a metal detector for characteristic railway features, can be used for a switch way detection, as speed or displacement sensor. Sensors such as cameras [14,15] or LIDAR [16,17] can directly identify the different switch ways and contribute to the trackselective result. A study of a tightly coupled localization with raw GNSS data and a track map was shown in [5,18].…”

“…Algorithm: Train Localization (RBPF) Input: GNSS and IMU sensor data Output: topological coord. ( , , ) and train speed (1) load map (2) initialize odometry Kalman filters with zero vector (3) initialize all particles by first GNSS position (30) (4) loop (5) if new measurement(s) available then (6) t i m es t e p : = + 1, Δ = − −1 (7) for all particles do (8) predict odometry KF (19) (9) update KF with speed (8)/acceleration (11) (10) if train is moving then (11) sample displacement from odometry (28) (12) compute map transition (21) (13) get geometry from map (train frame) (22) (14) compute likelihoods (9)/(10)/(14) (15) multiply particle weight by likelihoods (29) (16) else (train is stopped) (17) observe and filter gyroscope bias (18) end if (19) end for (20) normalize weights (27) (21) compute most likely output estimate (31)-(39) (22) if resampling necessary by eff then (23) perform resampling (24) end if (25) end if (26) end loop Algorithm 1: Algorithm of the map-based train localization with GNSS, IMU, and Rao-Blackwellized particle filter.…”

Rail detection using lidar sensors

“…2. Further properties, such as the spot size and beam divergence, or the combination of several lidar sensors go beyond the scope of this article and have been discussed in [15].…”

“…The larger the measurement rate and the smaller the train velocity are, the better the objects are discretized along the track (cf. [15,Fig. 3b]).…”

“…Besides others, level crossings [7], platforms [15], bridges and tunnels [6], masts of the overhead wiring as well as turnouts again can serve as landmarks. Our analysis of different sensor concepts discussed in [15] has shown that lidar (light detection and ranging) sensors have the largest potential to fill the remaining sensory gap for train-borne localization. Compared with camera systems, lidar sensors are independent of external (natural or artificial) illumination so that they can be operated every time and everywhere.…”