摘要
针对三维重建领域中,不同视角下点云的多视定位和配准效率问题,提出一种基于法向量改进的ICP算法。根据点云法向量间夹角特征选出关键点,计算关键点的曲率,通过主曲率特征获取初始对应点集,用高斯曲率和点间距离双重约束查找精确匹配点对,引入平衡因子的概念,给出适用范围,在不同的点云分布下,达到最优匹配,通过四元组法计算最优刚体变换。实验结果表明,相比传统ICP算法,改进后的算法将误差降低至0.05%,配准效率提高至70%以上,点云配准效率明显提升。
For the problem of point cloud visual orientation from different perspectives and registration efficiency in the three-dimensional reconstruction field,an improved ICP algorithm based on normal vector was proposed.Key points were selected according to the angle between the normal vectors of the point cloud.The main curvature of key points was computed to get the initial match points,and the dual constraints of Gaussian curvature and distance between points were used to select exact match points.The concepts of balance factor were introduced and the scope was given,the best match points then were selected in different point clouds.Quad method was used to calculate the optimal rigid transformation.Experimental results indicate that comparing with traditional ICP algorithm,the improved method shortens the registration error to 0.05% and raises the registration efficiency to over 70%.It is obviously that the registration efficiency is ascendant.
出处
《计算机工程与设计》
北大核心
2016年第1期169-173,共5页
Computer Engineering and Design
基金
总装预研基金项目(9140A17020113BQ04226)
关键词
点云配准
法向量
曲率
点间距离
平衡因子
point cloud registration
normal vector
curvature
distance between points
balance factor
