扩展高斯图像聚类的缺失点云配准

Registration of losing point cloud based on clustering extended Gaussian image

针对局部缺失点云配准时精度不高和收敛过慢等问题,提出了一种基于扩展高斯图像聚类的快速点云配准算法。通过将点云映射到扩展高斯图像中进行聚类后逆映射实际点云获取待配准的子点云,进而规避局部缺失带来的干扰;此外,为提高计算效率和配准精度,采用距离-曲率描述子查询对应点对进行奇异值分解进行粗配准,并结合迭代最近点精配准算法实现点云配准过程。实验结果表明,该算法对于局部缺失点云具有较高精度(均方误差相对于结合ICP的传统算子FPFH降低了近17.9%),且相比其它算法有一定的速度优势(耗时相对于结合ICP的SHOT算子加速了近32.5%)。该算法可以有效的运用在局部缺失点云的位姿识别中,从而可以被广泛地应用于工业现场中三维物体的快速识别定位。

With the aim of tackling the registration problems in terms of low matching accuracy and low convergence speed in locally losing point clouds,a fast point cloud registration algorithm based on a clustering extended Gaussian image is proposed herein.To avoid the interference due to local loss,the point cloud is mapped to the extended Gaussian image for clustering and inversely mapped back to the actual point cloud.Moreover,to improve the efficiency of computation and the accuracy of registration,the process of point cloud registration is realized by using the distance–curvature descriptor to obtain the corresponding point pairs and the iterative closest point(ICP)algorithm.The experimental results reveal that this algorithm displays high accuracy in the case of locally losing point clouds(resulting in a mean squared error(MSE)value lowered by 17.9%for the fast point feature histogram(FPFH)descriptor combined with the ICP algorithm).Moreover,it is faster than other algorithms(resulting in a decrease in running time by 32.5%for the signature of histograms of orientation(SHOT)descriptor combined with the ICP algorithm).Therefore,it can be widely applied for fast recognition and location of three-dimensional objects in the industrial field.