基于Kinect的三维重建关键技术研究

发布时间：2018-06-28 09:57

本文选题：Kinect + 三维重建　；参考：《天津理工大学》2017年硕士论文

【摘要】：三维重建(3D Reconstruction)是计算机视觉和图形学领域的一个重要研究方向。其主要目的是为物体或者场景建立可用于计算机处理和分析的三维模型,研究成果可应用于动画制作,虚拟现实,逆向工程,机器人导航等领域。微软公司生产的Kinect传感器为三维重建数据的获取提供了一种快捷方便的方式,有力的推动了三维重建技术的发展。基于Kinect数据的三维重建一般分为四个步骤:深度图像获取、深度图像转换、点集配准、点云拼接。点集配准是其中的关键步骤,点集配准的效率直接影响三维重建的精度。根据需要配准的目标的特性,点集配准可分为刚性配准和非刚性配准。本文分别针对刚性和非刚性的配准进行了研究。具体内容如下:1).提出了一种基于Spin-images和IRLS-ICP的深度图像刚性配准算法。ICP算法是当前刚性配准的主流算法,但ICP算法对初始对应关系敏感。针对这一问题,本文首先利用Spin-images特征对点集进行粗配准,找到相邻点云的大致对应关系,在此基础上再利用ICP算法进行精确配准。在公共数据集上的实验结果表明提出的方法可以很好的用于刚性配准。2).提出了一种基于IRLS-ICP-TPS的深度图像非刚性配准算法。对于可能发生形变的非刚性物体进行三维重建时,刚性的配准方法不能有效的配准发生形变的点云;因此本文在ICP算法基础上引入TPS算法来解决这一问题;首先使用IRLS-ICP找到两个待配准点集中的控制点,作为计算非刚性系数的基础。然后基于控制点,使用TPS计算出变换矩阵;最后将这个矩阵应用于其中一个点集,完成两个点集的配准。公共数据集和自制数据的实验结果表明提出的方法能够解决非刚性配准的问题并取得比较好的结果。
[Abstract]:Three-dimensional reconstruction (3D Reconstruction) is an important research direction in the field of computer vision and graphics. Its main purpose is to establish a three-dimensional model for computer processing and analysis for objects or scenes. The research results can be applied to animation production, virtual reality, reverse engineering, robot navigation and other fields. The K produced by Microsoft Corp Inect sensor provides a fast and convenient way to obtain 3D reconstruction data, which strongly promotes the development of 3D reconstruction technology. The 3D reconstruction based on Kinect data is generally divided into four steps: depth image acquisition, depth image conversion, point set registration, point cloud collage. Point set registration is the key step, point set registration Efficiency directly affects the precision of 3D reconstruction. According to the characteristics of the target that needs registration, the point set registration can be divided into rigid registration and non rigid registration. This paper studies the rigid and non rigid registration respectively. The specific contents are as follows: 1) a rigid registration algorithm based on Spin-images and IRLS-ICP is proposed,.ICP algorithm is proposed. The algorithm is the mainstream algorithm of current rigid registration, but the ICP algorithm is sensitive to the initial correspondence. In this paper, we first use the Spin-images feature to make rough registration of the point set, find the approximate relation of the adjacent point cloud, and then use the ICP algorithm to accurately match it. The experimental results on the public data set show the proposed method. The method can be used very well for rigid registration.2). A non rigid registration algorithm based on IRLS-ICP-TPS is proposed. The rigid registration method can not effectively register the deformed point cloud when the deformed non rigid object is reconstructed, and the TPS algorithm is introduced on the basis of ICP algorithm in this paper. In order to solve this problem, we first use IRLS-ICP to find two control points in the centralization of the registration points as the basis for calculating the non rigid coefficients. Then based on the control points, the transformation matrix is calculated by using TPS. Finally, the matrix is applied to one of the points set, and the registration of the two point sets is completed. The experimental results of the common data set and the self-made data show that the experimental results show that the results of the common data set and the self-made data show that The proposed method can solve the problem of non rigid registration and achieve relatively good results.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】