Application of 4D labelling in intelligent driving BEV perception

doi:10.13474/j.cnki.11-2246.2023.0279

Abstract

Abstract: Perception is an important part of intelligent driving application. With the advancement of BEV(bird's-eye view) perception algorithm in the industry, higher requirements are put forward for the quality and order of magnitude of data annotation. In this paper, the detailed requirements of BEV perception algorithm for data annotation are deeply studied, and a new 4D data annotation method upon vector space is proposed based on the weakness of current mainstream data annotation methods. The annotation results are projected into the image space in real time, and the output of massive training samples can be truly annotated once. Practical results show that 4D annotation can not only improve the annotation efficiency more than a hundred times, but also guarantee the quality of data annotation in essence and meet the needs of BEV perception.

Key words: BEV perception, 4D labelling, quality and efficiency, intelligent driving

CLC Number:

P237

ZHANG Panpan, SONG Jiashun. Application of 4D labelling in intelligent driving BEV perception[J]. Bulletin of Surveying and Mapping, 2023, 0(9): 139-143.

References

[1] 陈云培.浅析自动驾驶感知系统[J].汽车维护与修理,2022(15):74-75.
[2] CHENG Hong.Autonomous intelligent vehicles-theory,algorithms,and implementation[M].[S.l.]:Springer,2011.
[3] 王麒.基于深度学习的自动驾驶感知算法[D].杭州:浙江大学,2023.
[4] 蔡莉,王淑婷,刘俊晖,等.数据标注研究综述[J].软件学报,2020,31(2):302-320.
[5] LI Zhiqi,WANG Wenhai,LI Hongyang,et al.BEVFormer:learning bird's-eye-view representation from multi-camera images via spatiotemporal transformers[EB/OL].[2023-04-20].https://arxiv.org/abs/2203.17270.
[6] LIU Zhijian,TANG Haotian,AMINI A,et al.BEVFusion:multi-task multi-sensor fusion with unified bird's-eye view representation[EB/OL].[2023-04-20]. https://arxiv.org/abs/2205.13542.
[7] 王越,何伟,周琳,等.基于车载LiDAR数据的高精道路地图制作[J].地理空间信息,2022,20(6):92-95.
[8] 秦志伟.数据高效标注开启自动驾驶"进阶之路"[N].中国科学报,2021-07-22(3).
[9] 蒲建鑫,宋方伟,冷齐齐.基于SFM算法的三维重建关键技术研究[J].电子技术,2021,50(6):36-37.
[10] ZHU Siyu,ZHANG Runze,ZHOU Lei,et al.Very large-scale global SfM by distributed motion averaging[C]//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City:IEEE,2018:4568-4577.
[11] 李庆.基于深度学习方法的SFM三维重建技术研究[D].大连:大连理工大学,2019.
[12] ZHU Qingtian,MIN Chen,WEI Zizhuang,et al.Deep learning for multi-view stereo via plane sweep:a survey[EB/OL].[2023-04-20].https://arxiv.org/abs/2106.15328.
[13] 刘进.基于MVS点云的城市建筑三维重建应用研究[D].南宁:广西大学,2015.
[14] 魏珺洁.深度学习在图像自动标注中的应用初探[J].智能计算机与应用,2020,10(3):111-113.
[15] BAO Zhibin,HOSSAIN S,LANG Haoxiang,et al.High-definition map generation technologies for autonomous driving[EB/OL].[2023-04-20].https://arxiv.org/abs/2206.0540.