增加直线约束的网格化2D相机运动估计算法

2D mesh camera motion estimation algorithm with line preserving constraint

相机运动估计是视频稳像、视频拼接等技术的关键环节。网格变形模型可以在一定程度上解决传统的单个2D仿射或单应性变换模型难以处理的视差问题。在原CPW(Content Preserving Warping)算法的网格变形模型基础上,添加直线保持约束,并根据特征点的分布和检测出的直线结构特征,自适应地调整特征点约束、相似性约束和直线保持约束的权值。将三类约束均表示为二次项,利用稀疏线性系统求解得到网格化的相机运动路径。根据算法估计的相机运动对图像进行变形可以保持图像中人眼较为敏感的直线结构不被破坏,直观地提升了视觉效果,有效减少了图像变形带来的失真。

Camera motion estimation plays an important role in video stablization and stitching.To some extent,the mesh warping model can handle the problem of parallax,which is difficult for traditional 2D motion model with a single affine or homography matrix.A line preserving constraint is added to the CPW(Content Preserving Warping)algorithm,and the weight of the feature point constraint,similarity constraint and line preserving constraint are adaptively ajusted according to the distribution of feature points and the detected lines.Then three types of constraints are expressed as quadratic terms and the camera path is obtained by solving the sparse linear system.The algorithm keeps the linear structure,which is sensitive to human vision when warping image according to the estimated camera motion.As a result,the perceptual visual effect is improved,and the distortion caused by image warping is reduced.