复杂场景三维点云中未知球形目标的自动识别方法

Automatically Identifying Uncertain Spherical Targets in 3D Point Clouds of Complex Scenario

为了从散乱三维点云中识别并定位非确定个数、未知半径的球形目标,提出了复杂场景三维点云中未知球形目标的自动识别方法.首先利用局部邻域点集的二阶曲面逼近方法估计每个点的微分几何属性值;再根据各点的2个主曲率差异初步筛选疑似球面点,并根据法向和平均曲率推算各点对应的球心坐标;然后利用专门设计的层次聚类算法以球心坐标和估计半径为特征值对疑似球面点实施非监督聚类,将分属不同目标球的球面点区分开;最后逐类完成球面精确拟合.针对实验室布置场景的三维激光扫描点云进行实验的结果表明,该方法不仅具有较高的目标球定位精度,而且能够稳健地解决被障碍物遮挡情况下的非完整球面的半径识别和球心定位问题.

In order to identify uncertain number of spherical targets with different radius from terrestrial laser scanning point clouds, a method of automatically locating spheres and estimating their radii is proposed. Firstly, local second-order surface fitting of scattered measured point set is used to estimate the embedded differential geometric properties on the specified position of the measured surface. Then the two main curvatures of every point are compared, and all points on probable spheres are roughly identified according to the curvature difference point by point. At the same time the corresponding spherical center coordinates and radii are estimated. We specially design hierarchical clustering algorithm for estimated spherical centers and radii, which is utilized to separate probable points into distinct spherical targets. Finally, every category of spherical surface points is used to approximate spherical surface of its target whose accurate geometric parameters are computed subsequently. The scene demonstrate identify incomplete experimental results based on terrestrial laser scanning data of specially arranged that, this method can not only pinpoint the center of spheres, but also robustly spherical surfaces occluded by obstacles.