online read us now
Paper details
Number 4 - December 2022
Volume 32 - 2022
Vision-based positioning of electric buses for assisted docking to charging stations
Tomasz Nowak, Michał R. Nowicki, Piotr Skrzypczyński
Abstract
We present a novel approach to vision-based localization of electric city buses for assisted docking to a charging station.
The method assumes that the charging station is a known object, and employs a monocular camera system for positioning
upon carefully selected point features detected on the charging station. While the pose is estimated using a geometric
method and taking advantage of the known structure of the feature points, the detection of keypoints themselves and the
initial recognition of the charging station are accomplished using neural network models. We propose two novel neural
network architectures for the estimation of keypoints. Extensive experiments presented in the paper made it possible to
select the MRHKN architecture as the one that outperforms state-of-the-art keypoint detectors in the task considered, and
offers the best performance with respect to the estimated translation and rotation of the bus with a low-cost hardware setup
and minimal passive markers on the charging station.
Keywords
AI transport, localization, monocular vision, deep learning, keypoints, advanced driver assistance system