摘要
针对传统增量式三维重建方法重建点云较稀疏的问题,在稀疏重建算法的基础上,提出了一种序列影像稠密三维重建方法。首先,使用强制特征选择机制选取稠密特征,通过光流跟踪特征点获取稠密同名点对;其次,为保证同名点精度,利用前后向判决策略剔除外点,并利用稀疏重建获取的相机位姿,采用基于对极几何约束优化同名点坐标的方法对跟踪点坐标进行优化;最后,通过基于相机位姿的深度滤波器恢复稠密点云,实现稠密三维重建。分别使用标准数据集和实测数据集进行实验,结果表明:提出的算法可精确估计相机位姿,同时可获取稠密三维重建,能广泛应用于工程实践。
To solve the problem of sparse reconstruction in traditional incremental 3D reconstruction methods,a new method of dense 3D reconstruction from image sequences is proposed on the basis of sparse reconstruction.Firstly,forcing feature selection scheme is employed to extract dense features and optical flow method is utilized to obtain correspondences.A forward-backward error detection scheme is applied to reject outliers.By taking advantage of camera pose obtained from sparse reconstruction,the tracking points’coordinates are refined by the optimization method based on epipolar geometry constraint.Finally,a depth filter is designed to obtain dense reconstruction.Experiments are performed using benchmark data sets and measured data sets,respectively.The results show that the proposed method can obtain accurate cameras’poses and generate dense reconstruction,which can be widely used in engineering practice.
作者
王海燕
程传奇
WANG Haiyan;CHENG Chuanqi(Engineering University of PAP, Xi’an 710000, China)
出处
《信息工程大学学报》
2021年第6期663-669,共7页
Journal of Information Engineering University
基金
武警工程大学基础研究基金资助项目(WJ201924)。
关键词
三维重建
光束法平差
光流跟踪
前后向误差
深度滤波器
3D reconstruction
bundle adjustment
optical flow tracking
forward-backward error
depth filter