摘要：

视觉重建方法多数存在设备构成复杂、精度低、耗时高等问题。为此,提出一种单台手持式数码相机恢复场景的三维模型方法。采用经典的对极几何初始化相机位姿,使用改进PnP算法计算多帧图像位姿,通过选取新图像鲁棒特征点扩建场景,并运用基于局部关联帧的光束平差法优化摄像机位姿和三维点位置。实验结果表明,与传统的光束平差法相比,该方法能够有效节约时间,具有较强的鲁棒性。

关键词: 单目视觉, 三维重建, 关键帧, 摄像机位姿估计, 局部光束平差

Abstract:

Most of the visual reconstruction method have problems such as complicated equipment,low precision,and high time consuming.A three-dimensional model method for recovering a scene by a single handheld digital camera is proposed.The camera pose is initialized by the classic epipolar geometry.The improved PnP algorithm is used to calculate the pose in multi-frame images.By selecting new image robust feature points to expand the scene,the accuracy of the scene is improved,and the bundle adjustment method based on local correlation frames is used to optimize the camera pose and 3D point position.Experimental results show that compared with the traditional beam adjustment method,the method can save time effectively and has strong robustness.

Key words: monocular vision, 3D reconstruction, key frame, camera pose estimation, local bundle adjustment

中图分类号: 

• TP391