文物三维重建误差点的自动剔除方法

Automatic Error Point Removal Method for Reconstructed Cultural Relic Models

基于图像的文物三维重建方法具有设备成本低、场地限制小、可利用已有文物照片等优点,已逐步成为不可移动文物保护领域的数字化保护新手段.但由于文物图像采集受到背景干扰、结构复杂和光照不足等因素的影响,基于图像的文物重建点云通常带有大量的冗余和误差点.为此,提出一种文物三维重建点云误差点的自动剔除方法.首先,利用三维重投影方法计算三维点在图像集中的平均可见概率,将可见概率较低的误差点去除;接着,提出一种基于空间剖分的扩散聚类方法获取主体三维点云,去除与主体不连通的背景冗余点;在重建出三维网格模型后,再次通过三维重投影去除误差面片,最终参数化纹理得到逼真的文物三维模型.实验结果表明,该方法能够全自动地进行三维重建,在保留重建文物主体部分的同时去除大量冗余和误差点,为文物保护提供新的数字化技术支撑.

Image-based three dimensional reconstruction has the advantages of cheap equipment,no site limitation,and compatibility with existing images,thus has become a new way for digital protection of cultural relics. However,cultural relics’ images are often affected by background distraction,structural complexity and light insufficiency,the reconstructed point cloud always contains lots of redundant and error points. For this sake,we propose an automatic error point removal method for reconstructed cultural relic models in this paper. Firstly,it applies a reprojection method to compute 3D points’ average visibility probabilities of all images,and removes all of the points with low visibility probabilities. Secondly,we employ a clustering method based on space partition to filter the 3D point cloud,and use it to remove error points which are disconnected with the principal target object;after a 3D mesh model has been reconstructed,we apply the reprojection method again to remove error triangles,and parameterize the texture to obtain the final 3D model reconstruction. Plenty of experiments show that our method can reconstruct 3D cultural relic models automatically with a high fidelity and preserve the principal part of the model while remove most of noisy points or meshes. It thus provides a novel research strategy and technical supports for the digital protection of cultural relics.