Abstract: 3D reconstruction from uncalibrated image sequences is very important technology in computer vision and photogrammetry，the camera can be moved by hand and the motion need not be known，thus makes image data capture become easy and very convenient.We only get 3D point cloud reconstructed from the image sequence，so the paper puts forward a kind of 3D reconstruction method considering of image sequences combined with 3D models taking advantages of parallel lines and vertical lines in features extracting.The experiment shows that the combined method is much validated.

Key words: image sequences, 3D construction, multiple view geometry, RANdom SAmple Consensus（RANSAC） algorithm

摘要： 基于非标定序列影像的目标三维重建是一项非常重要的技术和研究热点，它使数据获取变得十分方便。基于影像序列的点匹配，得到的是一些点云，基于此，提出一个混合的三维重建方法：第一，通过物体三维点建立物体的数字形状模型（DSM）；第二，通过提取物体轮廓线，尤其是相互的平行直段和垂直线段，构建物体的轮廓线；第三，给合现存的三维数据模型，在目标显示和数据结构方面构建恢复三维物体。实验以一个茶筒为例，采用Java3D显示结果，取得良好的结果。

关键词: 序列影像, 三维重建, 多视几何, 随机采样一致性算法