An Energy-efficient and Fast KNN Search Accelerator for Large Scale Point Cloud Map

doi:10.1109/ICECS58634.2023.10382817

	An Energy-efficient and Fast KNN Search Accelerator for Large Scale Point Cloud Map
	Yunhao Hu1,2,3 ; Hao Sun1,2,3 ; Chunxu Guo1,2,3 ; Qi Deng1,2,3 ; Yajun Ha1,4
	2023-12-07
会议录名称	2023 30TH IEEE INTERNATIONAL CONFERENCE ON ELECTRONICS, CIRCUITS AND SYSTEMS (ICECS)
发表状态	已发表
DOI	10.1109/ICECS58634.2023.10382817
摘要	KNN (K-Nearest-Neighbors) search has been widely used in LiDAR (Light Detection and Ranging) localization and mapping algorithms in smart vehicles. However, the complex outdoor environment and the strict battery limitations of smart vehicles introduce a great challenge to develop an efficient KNN implementation for large scale point cloud maps. Unfortunately, existing KNN accelerators perform inefficiently in reducing search regions and transferring point cloud maps. To solve this issue, we propose a fast and energy-efficient KNN accelerator with two techniques. First, we propose a novel search technique (NSVS, nearest-sub-voxel-selection) to reduce the redundant search region based on the neighboring distribution (dense or sparse) of the query point in the search structure. Second, we design an adaptive data transfer technique to efficiently transfer point cloud maps with different data reuse ratio from external memory to accelerator via multi large-bit-width ports with random access mode or on-chip cache with sequential FIFO access mode. Experimental results show that our proposed KNN search accelerator achieves 9.1 times faster than state-of-the-art KNN implementations on FPGA. Moreover, energy efficiency results show that our proposed accelerator is 11.5 and 13.5 times higher than the state-of-the-art FPGA and GPU implementations. © 2023 IEEE.
会议录编者/会议主办者	Baykon Industrial Weighing Systems ; et al. ; IEEE ; IEEE Circuits and Systems Society (CAS) ; Isik University, Faculty of Engineering ; Savronik
关键词	K-nearest-neighbor search FPGA large scale point cloud map 3D LiDAR localization and mapping smart vehicles
会议名称	30th IEEE International Conference on Electronics, Circuits and Systems, ICECS 2023
会议地点	Istanbul, Turkiye
会议日期	4-7 Dec. 2023
URL	查看原文
收录类别	EI
语种	英语
出版者	Institute of Electrical and Electronics Engineers Inc.
EI入藏号	20240515478397
EI主题词	Field programmable gate arrays (FPGA)
EI分类号	525.2 Energy Conservation ; 716.2 Radar Systems and Equipment ; 721.2 Logic Elements ; 741.3 Optical Devices and Systems ; 921.5 Optimization Techniques
原始文献类型	Conference article (CA)
来源库	IEEE
文献类型	会议论文
条目标识符	https://kms.shanghaitech.edu.cn/handle/2MSLDSTB/345965
专题	信息科学与技术学院信息科学与技术学院_PI研究组_哈亚军组信息科学与技术学院_硕士生信息科学与技术学院_博士生
共同第一作者	Hao Sun
通讯作者	Yajun Ha
作者单位	1.School of Information Science and Technology, ShanghaiTech University, Shanghai, China 2.Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, Shanghai, China 3.University of Chinese Academy of Sciences, Beijing, China 4.Shanghai Engineering Research Center of Energy Efficient and Custom AI IC, Shanghai, China
第一作者单位	信息科学与技术学院
通讯作者单位	信息科学与技术学院
第一作者的第一单位	信息科学与技术学院
推荐引用方式 GB/T 7714	Yunhao Hu,Hao Sun,Chunxu Guo,et al. An Energy-efficient and Fast KNN Search Accelerator for Large Scale Point Cloud Map[C]//Baykon Industrial Weighing Systems, et al., IEEE, IEEE Circuits and Systems Society (CAS), Isik University, Faculty of Engineering, Savronik:Institute of Electrical and Electronics Engineers Inc.,2023.