改进PMVS三维重建点云滤波算法

AN IMPROVED POINTS CLOUD FILTERING 3D RECONSTRUCTION ALGORITHM BASED ON PMVS

针对在小范围场景进行单目视觉三维重建过程中,稠密点云模型存在大量离群点的现象,提出一种改进的点云滤波算法。将多视图稠密重建(Patch-based Multi-View Stereo,PMVS)算法与统计分析法相融合,对利用PMVS算法得到的稠密点云进行统计分析,设定标准距离并求解点云中每一个点到其所有邻近点的平均距离,去除平均距离大于标准距离的点。实验结果表明,融合后的点云滤波算法不仅剔除了大量离群点,还在保证目标物体细节特征的情况下对冗余的特征点进行一定程度上的消除,在提高重建表面真实度和精度的同时,为后期测量装配工作提供了可靠模型。

Aiming at the process of 3D reconstruction of monocular vision in small-scale scenes,a large number of outliers appeared in the dense point clouds model,this paper proposes an improved points cloud filtering algorithm to filter the outliers.It mixed together the PMVS algorithm and statistical analysis.The dense point clouds obtained by PMVS algorithm were statistically analyzed,the standard distance was set,the average distance in the point clouds from each point to all its adjacent points was solved,and the points whose average distance was greater than the standard distance were removed.The experimental results show that the fused points cloud filtering algorithm not only eliminates the outliers,but also eliminates redundant feature points that ensure the target details.It improves the accuracy and authenticity of the reconstructed surface,and provides a reliable model for later measurement and assembly work.