基于信息论的KL-Reg点云配准算法

Information Theory Based KL-Reg Point Cloud Registration

针对含有高噪声、体外点及不完整点云数据的配准失效问题,该文提出以信息论为理论基础,相对熵度量点云相似度的KL-Reg算法。该算法不需要显式地建立对应关系,首先将点云数据建模为高斯混合模型,然后用相对熵度量高斯混合模型间的分布距离,最后通过最小化分布距离计算模型变换。实验结果表明所提的KL-Reg算法配准精度高、稳定性强。

The registration of point clouds with high noises, outliers and missing data will be failure because the correspondence between point clouds is inaccurate. This paper proposes a information theory based point cloud registration method called KL-Reg algorithm without building correspondence. The method represents the point cloud with Gaussian mixture model, then computes the transformation through minimizing the KL divergence without build explicit correspondence. Experimental results show that KL-Reg algorithm is precise and stable.