未标定图像序列的隐式曲面重建方法

Reconstruction of Implicit Surfaces from Uncalibrated Camera Image Sequences

为了提高未标定图像序列三维重建得到的几何模型的质量,提出特征点检测算法,以得到更多的匹配点.其主要思想是在首帧图像指定密集的网格,在网格点附近确定最容易跟踪的特征点,利用迭代方法得到子像素精度的特征点坐标,然后用稀疏特征集的金字塔Lucas-Kanade光流跟踪算法跟踪这些特征点,再用自标定算法,重建出相对均匀和稠密的三维点云,最后利用基于径向基函数(RBF)的隐式曲面重建算法,生成目标的表面模型.多个图像序列的重建结果表明,本方法对纹理丰富的场景能够获得较好的重建结果.

To improve the quality of 3D geometric models reconstructed from uncalibrated image sequences, a feature tracking algorithm was proposed to match more points among a sequence of images. In the algorithm, a dense grid in the first frame is drawn, and easily tracked feature points near the grid points are determined. The sub-pixel coordinates of the feature points are found by iteration. The optical flow for this sparse feature set is calculated using iterative Lucas-Kanade method in pyramids. An even and dense 3D point-cloud is reconstructed with a self-calibration algorithm. Finally, an RBF （radial basis function ） implicit surface reconstruction was applied to generate the surface .model of the object. By experimenting with a number of image sequences, the reconstruction results show that the algorithm can obtain satisfactory surfaces for images with rich texture.