CoDRMA: Collaborative Depth Refinement via Dual-Mask and Dual-Attention for Bird’s Eye View Collaborative 3D Object Detection
Yang,Kang1; Wang, Yongcai1; Han, Yunjun2; Jia,Qingshan3
2024
Conference Name2024 20th IEEE Conference on Automation Science and Engineering
Conference Date2024年8月28
Conference PlaceBari,Italy
Abstract

In collaborative perception, camera-based approach is more informative and economical than Lidar-based approach. However, currently, camera-based methods still have a significant performance gap compared to the Lidar-based approach due to the difficulty and uncertainty involved in depth estimation. This paper introduces a strategy that refines depth estimation using foreground and background information, which empowers accurate Birds Eye View (BEV) collaborative 3D detection by multi-agents. Our strategy encompasses two stages: Initially, we introduce the Dual-Mask to enhance depth estimation and employ Birds Eye View (BEV) paradigms for integrating multi-viewpoint data, facilitating a comprehensive scene analysis. In the second stage, we generate pseudo-images by fusing depth and masks as auxiliary messages. A Dual- Attention scheme is proposed, which leverages multi-agent communication to augment auxiliary insights and further refine depth estimations. By refining the depth information twice, our method effectively improves BEV-based collaboration 3D object detection accuracy especially the occlused and long distance objects. Experiments on the OPV2V dataset show that our method achieves state-of-the-art performance in 3D object detection task among known camera-based methods, narrowing the gap with Lidar-based methods. Codes will be made available. 

Indexed ByEI
Sub direction classification决策智能理论与方法
planning direction of the national heavy laboratory虚实融合与迁移学习
Paper associated data
Document Type会议论文
Identifierhttp://ir.ia.ac.cn/handle/173211/57370
Collection多模态人工智能系统全国重点实验室_平行智能技术与系统团队
Corresponding AuthorHan, Yunjun
Affiliation1.School of Information, Renmin University of China
2.Institute of Automation, Chinese Academy of Science
3.Department of Automation, Tsinghua University
Recommended Citation
GB/T 7714
Yang,Kang,Wang, Yongcai,Han, Yunjun,等. CoDRMA: Collaborative Depth Refinement via Dual-Mask and Dual-Attention for Bird’s Eye View Collaborative 3D Object Detection[C],2024.
Files in This Item: Download All
File Name/Size DocType Version Access License
通信作者-CASE2024(重要会议)-(1601KB)会议论文 开放获取CC BY-NC-SAView
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[Yang,Kang]'s Articles
[Wang, Yongcai]'s Articles
[Han, Yunjun]'s Articles
Baidu academic
Similar articles in Baidu academic
[Yang,Kang]'s Articles
[Wang, Yongcai]'s Articles
[Han, Yunjun]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[Yang,Kang]'s Articles
[Wang, Yongcai]'s Articles
[Han, Yunjun]'s Articles
Terms of Use
No data!
Social Bookmark/Share
File name: 通信作者-CASE2024(重要会议)-CoDRMA_ Collaborative Depth Refinement via Dual-Mask and Dual-Attention for Bird’s Eye View Collaborative 3D Object Detection.pdf
Format: Adobe PDF
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.