Design of On-board Long-range Perception System for Train

Abstract

To address challenges posed by long braking distances, long perception distances, and difficulty in constructing clearance ahead of running trains in real time, a millimeter-wave radar-based long-range perception system is proposed to detect track boundaries and train contours. This system is equipped with a set of high-resolution long-range millimeter-wave radars that are adaptable to the strong scattering environment of rail transit and features an accurate method of calculating train speed based on the region of interest and the continuity hypothesis of speed. By comparing with the train speed, radar detection targets are classified into two categories: dynamic targets and static targets. Based on this classification, feature points of track clearance are extracted in a stable manner and boundary curve fitting is achieved using the seed method. In addition, an improved density-based spatial clustering of applications with noise (DBSCAN) method is proposed, using location and speed as clustering parameters and with different weights assigned to varying dimensions, to identify train contours by clustering and shape estimation based on the target track information output by radar, and assess collision risks according to the boundary information. Results from the on-board test demonstrate that the proposed system can realize stable fitting of the track boundary ahead and train contour within a range of 400 meters, which increases the detection distance by four times compared to that of traditional car radars, and enables collision warning to trains that may be involved in clearance invasion, thus helping avoid the occurrence of accidents.

Keywords

• track boundary detection
• density based clustering
• rail transit
• millimeter wave radar
• long-range perception