BEV-Based 3D Detection for Automatic Driving Using Lidar-Camera Fusion

Jihua Jiang

2024

Abstract

At present, deep learning technology and automatic driving related research are gradually mature. Autonomous driving perception technology has been developed tremendously as an important part of the autonomous driving system. This paper explores the development of a 3D detection task based on the Lidar-Camera fusion (LC Fusion) scheme of BEV technology from different fusion mechanisms. This paper concludes that the LC Fusion algorithm for BEV will be the most promising perception approach at present and the main form of perception system in the future. The BEV-based LC Fusion has many advantages such as high detection accuracy and robustness. Starting from the fusion granularity, this paper summarizes the characteristics of high accuracy and low latency of the current LC Fusion algorithm and the limitations such as network complexity, and proposes improvements such as attention fusion mechanism and network lightweighting for the problems faced by the algorithm. In addition, this paper proposes solutions such as unified spatial representation and decoupling of sensing channels, as well as the development direction of sensing systems including end-to-end design, multi-task learning, and knowledge distillation. This paper can provide reference materials and summarize perspectives for subsequent related researchers to pave the way for the development of this perception technology.

Paper Citation

in Harvard Style

Jiang J. (2024). BEV-Based 3D Detection for Automatic Driving Using Lidar-Camera Fusion. In Proceedings of the 1st International Conference on Data Science and Engineering - Volume 1: ICDSE; ISBN 978-989-758-690-3, SciTePress, pages 552-559. DOI: 10.5220/0012838400004547

in Bibtex Style

@conference{icdse24,
author={Jihua Jiang},
title={BEV-Based 3D Detection for Automatic Driving Using Lidar-Camera Fusion},
booktitle={Proceedings of the 1st International Conference on Data Science and Engineering - Volume 1: ICDSE},
year={2024},
pages={552-559},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012838400004547},
isbn={978-989-758-690-3},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Data Science and Engineering - Volume 1: ICDSE
TI - BEV-Based 3D Detection for Automatic Driving Using Lidar-Camera Fusion
SN - 978-989-758-690-3
AU - Jiang J.
PY - 2024
SP - 552
EP - 559
DO - 10.5220/0012838400004547
PB - SciTePress