Ego-Motion Compensation of Range-Beam-Doppler Radar Data for Object Detection
Abstract
With deep learning based perception tasks on radar input data gaining more attention for autonomous driving, the use of new data interfaces, specifically range-beam-doppler tensors, are explored to maximize the performance of corresponding algorithms. Surprisingly, in past publications, the Doppler information of this data has only played a minor role, even though velocity is considered a powerful feature. We investigate the hypothesis that the sensor ego-velocity, induced by the ego vehicle motion, increases the data generalization complexity of the range-beam-doppler data and propose a phase shift of the electromagnetic wave to normalize the data by compensating for the ego vehicle motion. We show its efficacy versus non-compensated data with an improvement of 8.7% average precision (AP) for object detection tasks.
Cite
Text
Meyer et al. "Ego-Motion Compensation of Range-Beam-Doppler Radar Data for Object Detection." European Conference on Computer Vision Workshops, 2022. doi:10.1007/978-3-031-25056-9_44Markdown
[Meyer et al. "Ego-Motion Compensation of Range-Beam-Doppler Radar Data for Object Detection." European Conference on Computer Vision Workshops, 2022.](https://mlanthology.org/eccvw/2022/meyer2022eccvw-egomotion/) doi:10.1007/978-3-031-25056-9_44BibTeX
@inproceedings{meyer2022eccvw-egomotion,
title = {{Ego-Motion Compensation of Range-Beam-Doppler Radar Data for Object Detection}},
author = {Meyer, Michael and Unzueta, Marc and Kuschk, Georg and Tomforde, Sven},
booktitle = {European Conference on Computer Vision Workshops},
year = {2022},
pages = {697-708},
doi = {10.1007/978-3-031-25056-9_44},
url = {https://mlanthology.org/eccvw/2022/meyer2022eccvw-egomotion/}
}