| Regulations for the registration of national cultural relics only require the collection of digital images for the digital records of characters,so that most of the movable cultural relics lack three-dimensional information.According to the industry standard for heritage conservation,the appearance characteristics which mainly includes the longest,widest and highest appearance parameters and the damage on the surface of three-dimensional cultural relics need to be recorded.Therefore,this thesis studies the 3D reconstruction of movable cultural relics and the extraction of their shape features.As for the 3D reconstruction,researcher mainly focus on the accuracy and efficiency of reconstruction.And for the contour feature extraction,researcher proposes the contactless high-precision extraction method.The main content of the study includes the following aspects.Firstly,the specific processes of reconstruction of 3D model of movable cultural relics based on multi-view geometry method is firstly to reconstruct spatial position of each view camera and sparse 3D point cloud of target object by incremental SFM algorithm,and then to reconstruct dense 3D point cloud based on CMVS/PMVS.In this study,researcher proposes to evaluate the correlation degree between images by the proportion of matching points,and studies the influence of the proportion of matching points on the reconstruction accuracy and efficiency.It is concluded that accurate and efficient reconstruction can be achieved when the proportion of matching points is about 50%.This conclusion can effectively guide the collection of image sequence based on multi-view geometric 3D reconstruction.In this study,the internal parameters of the imaging system are calibrated and the distortion of the image is corrected,so that the error of reconstruction angle is reduced from 0.023°to 0.015°,which improves the reconstruction accuracy effectively.Next,for the noise points of the reconstructed 3D point cloud model,the target point cloud is extracted from the environment by direct filtering,and then the color noise points are filtered out in the HSV color feature space.Finally,the detection algorithm based on standard deviation threshold is used to remove the outliers,which effectively improves the accuracy of the 3D point cloud model effectively and prepare for the next step of contour feature extraction based on 3D point cloud model.Finally,after obtaining the three-dimensional point cloud model,the directed bounding box algorithm is used to calculate the longest,widest and full-highest of the movable cultural relics.In view of the difference in density distribution of the point cloud model,the weighted principal component analysis method based on the density degree of the point cloud is proposed to calculate the main direction of the directed bounding box.The reproducibility standard deviation of dimension characteristics of the Yuan dynasty porcelain vase measured by the proposed method is less than 0.15mm,and the maximum measurement error is less than 0.25mm,which is higher than the requirements of shape parameters measurement in the protection and research of movable cultural relics.Using point cloud segmentation algorithm based on region growth of color information to detect cultural relic surface damage can realize three-dimensional detection and localization of cultural relic surface damage in the whole. |
