动态扫描场景下GM-APD激光雷达点云高精度配准方法研究

打开文本图片集

中图分类号：TN958.98; 文献标志码：A doi：10.37188/CO.2025-0073 CSTR：32171.14.CO.2025-0073

High-precision registration methods for GM-APD LiDAR point clouds in dynamic scanning scenarios

ZHONG Guo-shun1*，LIU Qiu-zuo1，LI Meng1，PENG Tao²， SUN Jian-feng²，LIU Jian-wei1 （1.The1lth Research Institute ofChina Electronics Technology Group Corporation，Beijing 10o015，China; 2. National Key Laboratory ofLaser Spatial Information， Harbin Institute of Technology， Harbin 150001，China） * Corresponding author，E-mail： james830411@163.com

Abstract： Aiming at the issues of low overlapping rates of adjacent-frame point clouds and the tendency to forcibly register non-matching point pairs in Geiger-mode avalanche photodiode （GM-APD） LiDAR when used in dynamic scanning conditions， an enhanced iterative closest point （ICP） algorithm is proposed based on a bidirectional matching scheme and multi-resolution neighborhood expansion to improve registration accuracy and robustness. First， a K-D tree-based bidirectional search identifies overlapping regions between consecutive frames， enabling accurate initial alignment. Then， a high-resolution neighborhood expansion approach，weighted by local curvature similarity，is applied to refine the transformation matrix and suppress mismatched correspondences. Finally，a cascaded compensation mechanism ensures global consistency across frames.Experiment results demonstrate that our method achieves average distance errors of 0.21m （ 2km scene）and 0.10m （ 400m scene），effectively improving registration precision in dynamic scenarios and offering valuable support for 3D reconstruction.

Key Words：GM-APD；point cloud registration;pose correction;bidirectional matching；multi-resolution neighborhood expansion

1引言

2动态点云配准原理

盖革雪崩光电二极管（GM-APD）激光雷达凭借其单光子级探测灵敏度和纳秒级时间分辨率，在远距离探测中展现出显著优势，成为远距离三维成像的重要技术手段。（剩余12988字）