Research On Registration Method Of 3D Point Set Surfaces

In three-dimensional space,surface matching is the core technology of reverse engineering and artificial intelligence vision,and it is also the basic work of application fields such as virtual reality,industrial production,cultural relics repair,and medical image registration.Currently,the three-dimensional surface matching methods based on the point clouds which are scanned by laser device are mostly based on the information including point cloud coordinates,color and intensity,which imposes limitations on the universality of the method.Due to the essence of spatial surface registration is the matching of surface geometry,for this reason,this paper will do more in-depth research of surface matching with the point cloud that only contains vertex coordinate information.The main work is summarized as follows:(1)An expansion algorithm of three-dimensional key point extraction which is suitable for surface matching is designed.For the point cloud data that only contains vertex coordinate information processed in this paper,the curvature information of each vertex in the point cloud data is calculated as a substitute value of the point cloud intensity information or color information required in the scale-invariant feature transform algorithm.Improving the applicability of scale-invariant feature transform algorithms.Finally,the key points are extracted by the improved scale-invariant feature transform algorithm to obtain a set of key points.(2)A key point matching algorithm based on fast point feature histograms and model main trend constraint is given.Firstly,the fast point feature histogram at key points is calculated.According to the fast point feature histogram at the key point,the corresponding relationship between the key points is preliminarily filtered.Then,the main trend of the model is calculated by principal component analysis,the angle between the vector from the model center to the key points and the vector of main trend is used as a constraint condition to further screen the correspondence between key points.Finally,according to the constraint condition of the feature description at the key point and the model main trend vector,the more accurate set of the key point pair is calculated.(3)A method for solving the rigid transformation parameters based on the set of key point pair is given.Firstly,calculate the Euclidean metrics from the key point to respectively model centers for each group of key point pairs;then,calculate the ratio of the Euclidean metrics between the key point pairs;then,obtain the scaling factor between the two cloud models by calculate the average of all ratio of Euclidean metrics of the key point pairs to achieve the scale matching of the two point cloud model;finally,rotational translation parameters of the cloud model are solved by using the singular value decomposition method based on the key point pairs,to achieve a coarse matching of the model surface.(4)An iterative optimization method of surface matching is implemented.Firstly,the initial position and pose estimation between the two model surfaces is obtained through the precise set of key point pair.Then,the second matching based on the ICP algorithm is performed to complete the accurate matching between the two model surfaces.