中文摘要：

基于图像中数码相机嵌入的元数据信息，提出了一种可变内参数的序列图像重构算法，首先从序列中选择两幅图像建立初始结构，然后依次将其他图像加入当前重构结果，进一步通过集束调整来最小化序列中所有图像重投影误差，得到精确的三维重构结果，避免了复杂繁琐的自标定过程。实验结果验证了算法的有效性。

英文摘要：

This paper proposed a method for 3D scene reconstruction from image sequence. The algorithm could deal with varying internal parameters sequence. Based on metadata information embedded by digital camera, it first selected a pair of views to setup the initial structure and motion, and then added other views in sequence to update the existing structure and motion. Bundle adjustment was used to minimize re-projection error across all images of the sequence. The algorithm avoided complex self-calibration procedure. Results on real datasets demonstrate the feasibility and accuracy of the method.