【作者】 陈欣；

【导师】 高隽；

【作者基本信息】 合肥工业大学 ， 信号与信息处理， 2016， 硕士

【摘要】 基于视觉的位姿测量是当前计算机视觉中的研究热点,而基于光学传感器的位姿测量方法因为无需接触目标就能直接获取目标位姿信息,在国防、航天航空、工业、医学等各个领域得到广泛应用。本文主要研究基于计算机视觉的非合作目标的位姿测量方法,针对基于单目视觉圆位姿的二义性问题,设计了一种融合深度图像和彩色图像的圆位姿测量算法,实验验证了该算法的可行性和有效性。本文主要研究内容如下：(1)介绍了当前基于视觉的非合作目标的位姿测量的研究现状,阐述PMD相机的测距原理、数学模型、成像模型。研究了PMD相机的标定方法,并完成了PMD相机的标定实验。通过PMD相机和RGB相机间的平移旋转变换关系,完成了深度图像和彩色图像的立体配准。(2)研究并分析了椭圆检测的方法,实现了一种基于弧段的椭圆检测。在弧段组合过程中,设计一种联合凸性和距离夹角的新的弧段组合方法,有效减少了弧段组合中不必要弧段,降低了椭圆检测的失误率。(3)针对基于圆的单目视觉位姿测量中解的二义性问题,融合深度信息和高分辨率的彩色图像信息,形成高分辨率的深度图像,获取目标的三维点云,剔除传统基于圆的单目视觉位姿测量中的虚假解。通过搭建实验环境,采集实验数据,计算位姿并分析实验结果。更多还原

【Abstract】 Vision-based pose measurement is a hot topic in the computer vision, and the position and attitude measurement based on the optical sensor can acquire the target’s position and orientation information without touching the target directly. Now it is widely used in national defense, aerospace, industrial, medical, and other areas. I mainly focus on the pose measurement of non-cooperative target based on computer vision research. In order to solve the ambiguity problem of circle pose based on monocular vision, we designed a circle pose measurement algorithm which is base on the fusion of depth image and color image. The experiment verification the feasibility and effectiveness of the algorithm.This thesis includes the following contents:(1) Have an introduction about pose measurement of non-cooperative target based on vision and the basic measurement principle, the math model and imaging model. We study the calibration of the PMD camera, and complete the PMD camera calibration experiments. By using the rotation and translation between PMD and RGB camera, we implement the registration between depth image and color image.(2) The methods of ellipse detection were researched and analyzed. We realize ellipse detection based on arcs combination. In the process of arcs combination, we design a new arc combination method by jointing distance, angle and convexity between arcs. It effectively reduces the unnecessary arc of arc combination, and the error of ellipse detection.(3)Due to the ambiguity of circle pose estimation based on monocular vision, we obtain a high-resolution depth image by the fusion of color image and depth image, and then we get 3D point cloud of target. By using these information, we can remove the false solution of circle pose estimation based on monocular vision. By building the experimental environment and collecting experimental data, we calculation and analysis of experimental results pose.更多还原