结合图像信息的快速点云拼接算法

现有三维激光扫描设备通常配有一个同轴相机,它可以对扫描场景进行拍摄。针对带有同轴相机的激光扫描设备,本文提出了一种结合图像信息的快速点云拼接算法。与传统拼接算法同时计算点云间的旋转和平移变换不同,本文对这两种变换分别进行求解。其中,不同扫描点云间的旋转变换是利用视觉几何知识由同轴相机在不同扫描站点下拍摄的图像直接获得,而平移变换是由本文提出的改进ICP算法得到。在改进的ICP算法中,只有平移变换的3个未知量被迭代计算,其输入是去除旋转变换后的点云。试验结果表明利用图像获得的点云旋转变换具有很高的准确性;并且由于本文算法中迭代过程只针对平移变换的3个变量进行计算,因此与需要迭代计算6个变量的传统ICP算法相比,本文算法计算复杂度大幅降低,同时更易收敛于全局最优值且收敛速度有所提高。

Fast Registration Method for Point Clouds Using the Image Information

On the existing laser scanners, there usually is a coaxial camera, which could capture images in the scanning site. For the laser scanners with a coaxial camera, we propose a fast registration method using the image information. Unlike the traditional registration methods that computing the rotation and translation simultaneously, our method calculates them individually. The rotation transformation between the point clouds is obtained by the knowledge of the vision geometry and the image information, while their translation is acquired by our improved ICP algorithm. In the improved ICP algorithm, only the translation vector is updated iteratively, whose input is the point clouds that removing the rotation transformation. Experimental results show that the rotation matrix obtained by the images has a high accuracy. In addition, compared with the traditional ICP algorithm, our algorithm converges faster and is easier to fall into the global optimum.