Registration Of 3D Point Cloud Model And CT Slice Model For Cultural Relics

With the development of the days,the management and preservation of cultural relics have gradually received attention,and digital research has also become a hot spot.The common measurement method is to use a three-dimensional laser scanner to obtain the surface data of cultural relics,but the internal damage cannot be found in time.The application of CT technology makes up for this defect and provides a scientific basis for the actual restoration of cultural relics.Therefore,more accurate modeling of CT slice images is necessary.In this paper,the model registration technology is used to register the point cloud model and the CT slice model,and the error is analyzed to realize the evaluation of the CT slice model.The model registration technology is applied to the image modeling based on CT slices,and the optimal image segmentation threshold is found on this basis,so as to better guide the reconstruction of the model.The contents of this study are as follows:(1)Aiming at the problem that the number of data points in the CT slice model is too large,the efficiency and accuracy of the subsequent model registration are poor,and a method for quickly selecting contour feature points is proposed.Firstly,the points with small curvature are removed by the circular curvature method.Then the polygon approximation method is used to further extract the contour feature points.Finally,the contour feature points of the extracted model are accelerated by p Threads parallel calculation method,and the acceleration ratio is about 2.5 times.Compared with the uniform selection method and the polygon approximation method,this method overcomes the disadvantage of losing some features,effectively extracts representative features from slice images,and reduces data redundancy.(2)Aiming at the low efficiency and accuracy of registration using the original data point set,a model registration algorithm based on contour feature points is proposed.First,calculate the eight vertices of the minimum bounding box of the point cloud model of the same object and the contour feature point set of the CT slice model,select three non-collinear corresponding points,and achieve the initial registration by aligning the three pairs of reference points,effectively and quickly reducing the positional deviation of the two models,as prepare for fine registration.Then,the improved ICP algorithm is used in the fine registration stage.The method is to find the corresponding point pairs between the two models by constructing a two-way kd-tree,and then use the median distance method to remove the misregistered point pairs.Finally,the ICP algorithm is used to achieve fine registration,and calculate the registration time and error values.Compared with ICP algorithm,the method in this paper improves the accuracy by about 45%,and reduces the algorithm time by about 40%.(3)Aiming at the problem that the image segmentation threshold will affect the modeling accuracy of CT slice images,a point cloud model-guided CT slice image modeling method is proposed.When modeling based on CT slice image,the most important thing is to extract the high-precision contour of the image,and whether the image can be accurately segmented has a significant impact on the contour extraction,which is the basis of 3D reconstruction.The method first uses a laser scanner to model the surface of cultural relics in 3D to obtain an accurate point cloud model,and then registers the point cloud model with the contour feature point sets of the CT slice images extracted under different image segmentation threshold conditions,and obtains the corresponding error value.The threshold corresponding to the smallest error value is the best threshold for image segmentation.Under this threshold,the obtained 3D modeling results are the most accurate.Taking the vase model as an example,when the threshold is 166,the image segmentation effect is the best.