单Kinect+回转台的全视角三维重建

为了解决当前全视角三维扫描系统价格昂贵操作复杂的问题，提出利用1台Kinect和1个回转台来构建全视角三维模型的方法。研究涉及点云预处理、点云配准、全局误差修正以及色差修正等技术。首先使用Kinect采集数据并预处理，结合转台约束利用图像几何特征进行粗配准，随后使用迭代最近点（Iterative closest point，ICP）算法实现点云的精确配准。对于累积误差导致的闭环问题以及不同角度拍摄引起的色差问题，通过全局误差修正与色差修正算法处理，提升重建结果的精度。实验结果表明：该方法可以实现三维物体的全视角重建，并在精度上优于微软的KinectFusion方法。

Single Kinect and Rotating Platform for Full-View 3D Reconstruction

Traditional full-view 3D reconstruction systems are expensive and complex. To solve the above problems, a new algorithm for reconstructing an omnidirectional 3D scanning system is proposed based on a single Kinect and a rotating platform. The algorithm involves point cloud preprocessing, registration, minimizing global error between coordinate frames and color rectification. Firstly, RGB-D data are acquired by a Kinect sensor and preprocessed, and then a matching method is developed under the swivel table constraint for coarse registration. Further, the iterative closest point (ICP) algorithm is utilized for fine registration. Finally, for the loop closure caused by cumulative error and the color difference induced by different shooting angles, a global error correction and a color correction algorithm are conducted to improve the accuracy of the reconstruction results. Experimental results show that the reconstruction method can achieve full-view reconstruction of 3D objects and is superior to the KinectFusion method of Microsoft in accuracy.