基于深度传感器的多视角点云配准研究
Research on registration of multi-view point cloud based on depth sensor
刘耀文 1毕远伟 1张鲁建 1黄延森1
作者信息
- 1. 烟台大学计算机与控制工程学院,山东烟台 264005
- 折叠
摘要
为了解决大尺寸对称物体在多视角配准过程中出现的误匹配点对和累计误差问题,提出了 一种基于深度传感器的多视角点云配准算法.首先,使用深度传感器获取目标物体不同视角下的多片点云并进行预处理,对物体单侧相邻点云采用超四点快速鲁棒匹配算法(Super 4-points congruent sets,Super4PCS)进行粗配准,利用改进的点到平面ICP算法去除误匹配点对并进行精配准,之后将左右两部分的点云拼接,从而获取完整的三维点云模型.最后,针对多视角配准出现的累计误差问题,提出了一种全局优化方法从而减少累计误差.实验结果证明所提方法可以精准地完成多视角点云配准,获得准确的三维点云模型.
Abstract
To address the issues of mismatched point pairs and cumulative errors encountered during the multi-view point registration of large-scale symmetrical objects,a multi-view point cloud registration algorithm based on depth sensor is proposed.Firstly,the proposed approach leverages depth sensors to capture multiple point clouds of the target object from various viewpoints,which are then subjected to a series of preprocessing steps.To achieve coarse registration,the Super 4-points congruent sets(Super4PCS)is employed specifically for adjacent point clouds on one side of the object.Subsequently,an enhanced point-to-plane ICP algorithm is utilized to refine the registration by e-liminating erroneous point pairs.The resulting refined point clouds from the left and right sides are seamlessly com-bined,thereby generating a comprehensive 3D point cloud model.Furthermore,to mitigate the issue of cumulative er-rors arising from the multi-view registration process,a global optimization technique is introduced.Experimental eval-uations demonstrate the effectiveness and accuracy of the proposed method in achieving precise multi-view point cloud registration and generating highly accurate 3D point cloud models.
关键词
深度传感器/累计误差/多视角/点云配准Key words
depth sensor/cumulative error/multi-view/point cloud registration引用本文复制引用
出版年
2024
NETL
NSTL
万方数据