Three dimensional (3D) reconstruction is an important research field in computer vision. More and more attentions have been paid on urban scene reconstruction recently with development of digital earth, 3D city map etc. In this dissertation, the main work is focused on merging the structured information, such as lines and planes, with the traditional texture-based point reconstruction method, to enhance the accuracy and robustness of urban scene reconstruction. The main work is as follows: 1.Propose a method for indirectly using line features in 3D reconstruction, avoiding the drawbacks of directly using line features in structure recovery. Feature points are sampled and matched on each matched line pair. And their corresponding 3D points are accordingly adjusted to meet the collinear constraint after they have been reconstructed. 2.Performing color image segmentation to get coplanar information from images, and utilize this information to enhance the initial reconstruction of 3D scenes. 3.Propose a PCA-based method to integrate the above structure refinements into a statistical inference process, combining multiple view geometry with structure information extracted from color images, to improve the effectiveness and robustness of the structure refinement.
修改评论