特征检测跟踪与步场景生成算法研究.pdfVIP

摘 要 基于视觉的相机跟踪技术，是计算机视觉领域中的一个重要课题。主要就是 通过图像分析处理和计算机视觉的相关方法，实现对相机在真实场景中位置和方 向的跟踪。 本文研究了一种在未知场景中估计相机姿态，同时生成同步场景的方法，该 方法是基于自然特征的检测与跟踪实现的。该方法所实现的系统可用于小范围场 景中便携式相机运动的跟踪。 本文的主要研究内容如下：(1)对相机实时采集的图像进行预处理，而后进行 FAST-9 角点特征检测跟踪，得到两帧选定图像匹配点对的集合。(2)基于图像匹配 点对的集合，由 5 点算法估计初始的相机姿态。(3)根据相机姿态和相机内参，对 匹配点进行三角测量，得到图像匹配点对所对应的3D 空间点，并通过极线搜索增 加3D 空间点。(4) 由3D 空间点的集合估计显著平面，调整相机姿态与3D 空间点 坐标。(5)将3D 空间点投影到当前帧，通过计算并最小化重投影误差更新相机姿态。 (6)生成同步场景，并借助OpenGL 予以显示。 关键词：特征检测跟踪 相机姿态 显著平面 同步场景 Abstract The vision-based camera tracking is an important subject of Computer Vision. It tracks the location and orientation of a camera in real scene mainly by image analysis and some relevant methods of Computer Vision. This paper researches on a method of estimating camera pose and mapping the simultaneous scene in an unknown scene. This method is based on the detection and tracking of natural features. The result system can be used to track a hand-held camera in a small workspace. The main contents are as follows: (1)Get a set of feature correspondences by the detection and tracking of the FAST-9 corners. The FAST-9 corners are in frames which were captured by the camera and have been pre-processed. (2)Employ the five-point stereo algorithm to estimate the initial camera pose with the set of feature correspondences. (3)Triangulate the feature correspondences to get the corresponding 3D map points, according to the camera pose and internal parameters. And then add more map points by epipolar search. (4)Calculate a dominant plane according to the set of 3D map points. Then adjust the camera pose and the coordinates of these map points. (5)Reproject the 3D map points into the current frame, and then update the camera pose by minimizing an