A Mesh Reconstruction Algorithm Based On A Sequence Of Images

3D reconstruction is an important embranchment of CG, and is widely used today. So far, there is no general technology to solve the problem. Our system, introduced in this paper, is oriented at cheaper home use, and provides ideal low complexity and strong robustness. It uses a sequence of images and related DLT parameters of each photo to produce a 3D mesh.The reconstruction system can be roughly divided into 3 parts, which are camera calibration module, mask extraction module and mesh reconstruction module, according to the sequence of execution. To reconstruct a mesh, we propose a ray-tracing algorithm to calculate the 3D surface points, which then follows a mesh extraction method based on the cross-section contours.This paper will describe the algorithms, key technologies and our enhancement of the reconstruction module in details. A so-called sandwich test is invented to reduce the complexity. The idea of intervals and interval sets is introduced so that we can deal with the culling problem among surface points by an intersection operation of several interval sets. We solve the "branching structure" problem by adding a directed edge pairs. A dynamic programming technique is used in the mesh extraction process to make the algorithm more efficient. Some 3D models are reconstructed in our system, which proves the efficiency and robustness of all algorithms.The calibration module in this system uses the DLT parameters. A calibration approach that requires only a small amount of data is proposed for some extreme case. By the end of our article, we realize the limitation of our algorithm and point out some possible solutions in future work.