Feasibility Study of Vision-based Localization Method for Stopping Control Using Real Environment Data

摘要

Stopping control is very important in public transportation systems. It is used to precisely match where a vehicle is boarded at a station and the positions of the doors of the vehicle. Although it is necessary to estimate self-localization with high accuracy, the cost of implementing it using balises, which is currently popular, is very high. To accelerate the introduction of stopping control to public transportation systems, a low-cost self-localization estimation technology is required. We propose a vision-based self-localization method for underground railway, which does not use expensive balises as that in conventional method. The method uses image recognition to calculate the relative distance to a ground target, whose position is known, and estimates the position of a vehicle from the absolute position of the target and the relative distance. In this paper, we evaluate the accuracy of the self-localization estimation using the proposed method under outdoor conditions that are more severe compared with underground conditions. From the results of six trials, we confirmed that the accuracy of position estimation of all trials satisfied the required accuracy of 0.8m, when the distance between the vehicle and ground target was 20m.