三维重建中先验概率重加权点云配准算法  

Point Cloud Registration Algorithm for 3D Reconstruction Based on Prior Probabilistic Weighting

作　　者：孙志亮 张荣国[1] 赵建[1] 李富萍[1] 胡静[1] SUN Zhi-liang;ZHANG Rong-guo;ZHAO Jian;LI Fu-ping;HU Jing(School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学计算机科学与技术学院,太原030024

出　　处：《太原科技大学学报》2021年第5期367-373,共7页Journal of Taiyuan University of Science and Technology

基　　金：国家自然基金(51375132);山西省自然科学基金(201801D121134);太原科技大学博士科研启动基金(20202057)。

摘　　要：基于高斯混合模型的三维点云配准算法在面对大量的外点及对应位置缺失时表现不佳。为此提出了一种结合点到面距离和先验概率重加权的点云配准方法。首先,通过高斯混合模型和均匀分布建立点云之间的位置对应关系;其次,使用先验概率对高斯混合模型的混合比例重新加权来处理对应位置缺失,同时利用后验概率推测潜在外点及其比率;然后,向误差函数中添加目标点的法向,用点到面距离度量点云之间的相似性;最后,在求解高斯混合模型参数时移除潜在外点来提高算法准确性。在合成数据和真实场景中进行实验并采用不同指标进行评估,结果验证了本文方法的有效性。The existing 3D point cloud registration algorithms based on Gaussian Mixture Model(GMM)perform poorly when facing with vast outliers and missing correspondence.A point cloud registration algorithm that combines the point-to-plane distance and the weighted prior probability is proposed.Firstly,the Gaussian mixture model and uniform distribution are used to establish the correspondences.Then,the prior probability is utilized to reweight the GMM mixture ratio to deal with missing correspondences;meanwhile,the posterior probability is applied to infer potential outliers and their proportion.Besides,the point-to-plane distance is introduced to measure the similarity between point clouds by adding the normal direction of the target point to the error function.Finally,potential outliers are removed to improve the accuracy of the algorithm when solving the GMM parameters.The experimental and evaluation results prove the effectiveness of the method on synthetic data and real scenes.

关 键 词：点云配准 高斯混合模型 先验概率 潜在外点 对应位置缺失 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]