Curvature Classification for Trains using Along-Track and Cross-Track Accelerometer and a Heading Rate Gyroscope

Abstract

Real time curvature classification is crucial for all train localization problems. A reliable method to detected the track taken by the train after a switch is necessary and essential for train collision avoidance systems. At a larger scale, this should be included in a global surveillance system. In this paper, we discuss three possible track curvature detection methods based on two accelerometers and one gyroscope. We define and analyze corresponding test statistics that determine the actual track curvature. Given system safety requirements, i.e., maximum allowable probabilities of false alert and miss-detection, we derive a minimum detectable curvature difference (MDCD) between two possible tracks and compare these values with standard curvatures used in Germany. In this paper, it is shown that these MDCDs strongly depend on the sensor quality (for which an analytical form of the Gaussian error overbound of the sensor error is derived) and on the train dynamics (velocity). This analysis shows for two detectors very promising results and suggests a possible optimal combination of their test statistics.