基于ORB特征的单目视觉定位算法研究

针对大范围室外场景和具有重复、高频纹理特征(例如水泥地、草坪)的场景,提出了一种鲁棒性强、定位精度高、速度更快的视觉定位算法。采用8级图像金字塔的ORB (Oriented FAST and Rotated BRIEF)特征描述子提取图像特征点,通过K近邻(KNN)匹配相邻图像序列的特征点对,依次解算基础矩阵F和本质矩阵E,采用自适应法利用单应矩阵和本质矩阵进行位姿估计,最后解算两帧图像间相机刚体运动的旋转R和平移t,利用三角测量法则求解出匹配点的三维坐标,重建相机运动轨迹。为了提高算法性能,提出采用最小化基于点特征的非线性重投影误差优化三维点。通过调用OpenCV在C++中实现,对所采集的数据集进行测试,测试结果表明,该方法比传统的3D位姿估计更优,实时可行。由于其基于单目而实现,因此无法得到尺度信息。

Monocular Vision Alignment Algorithm Based on ORB

We proposed a monocular visual location algorithm which is faster,higher precise,and more robustness than current state-of-the-art methods in a wide range,repeated,high-frequency texture features (for example,cement,lawn) scenes.We extracted ORB (oriented fast and rotated brief) feature of sequences,matched point matches of images by using KNN,then computed the fundamental matrix F and the essential matrix E and used them to compute the initial camera pose.Then we computed the rotation matrix R and the translation vector t for every pair of images,and computed the camera motion between the current and the previous image,triangulating 3-D points (structure) from 2-D image correspondences,tracking the trajectory of camera.In order to improve our algorithm,we refined the pose and structure through minimizing the reprojection error.A test was made on our datasets which has been implemented in OpenCV/C++.The results is much better than tradition algorithm.Since it is a monocular implementation,we cannot do absolute scale estimation.