融合点、对象、关键点等3种基元的点云滤波方法

Filtering of Point Clouds Using Fusion of Three Types of Primitives Including Points,Objects and Key Points

基元是影响点云滤波精度和效率的关键因素之一。本文提出了一种基于多基元的三角网渐进加密(MPTPD)滤波方法。它包括点云分割、对象关键点提取、基于关键点的对象类别判别3个主要阶段,且3个阶段的基元分别为点、对象、关键点。使用了4景机载激光雷达和摄影测量点云数据对MPTPD、三角网渐进加密(TPD)、基于对象的三角网渐进加密(OTPD)3种滤波方法进行了性能测试。试验表明,MPTPD方法具有整体上最优的性能:在精度方面,MPTPD与OTPD两种方法的精度相当,MPTPD方法的一类误差I、总误差T比TPD的相应误差分别低约22.07%和8.44%;在效率方面,多数情况下TPD、MPTPD、OTPD方法的效率依次降低,且MPTPD的平均耗时是OTPD平均耗时的57.93%。

Primitive, being the basic processing unit, is one of the key factors to determine the accuracy and efficiency of point cloud filtering. Triangular irregular network （TIN） progressive densification （TPD） and object-based TIN progressive densification （OTPD） are two existing filtering methods, but single primitive is employed by them. A multiple-primitives-based TIN progressive densification （MPTPD） filtering method is proposed. It is composed of three key stages, including point cloud segmentation, extraction of key points of objects, the key-points-based judging of the objects. Specifically, point, object and the key points are the primitive of the above three stages respectively. Four testing datasets, including two airborne LiDAR and two photogrammetric point clouds, are used to verify the overall performances of the above three filtering methods. Experimental results suggest that the proposed MPTPD has the best overall performance. In the viewpoint of accuracy, MPTPD and OTPD have the similar accuracy. Moreover, compared with the TPD, MPTPD is able to reduce omission errors and total errors by 22.07% and 8.44% respectively. In the viewpoint of efficiency, under most of the cases, TPD is the highest, MPTPD is the second, and OTPD is the slowest. Moreover, the total time cost of MPTPD is only 57.93% of the one of OTPD.