Research On Algorithm Of 3D Registration Of Complex Point Cloud

With the development and popularization of optical technology in recent years,it has become very easy to obtain point cloud data with high accuracy.However,there are still some complicated objects with complex structures and surfaces which lead to incompletely scan data.Therefore,point cloud registration technology is required to process the point cloud data obtained from various angles of view to construct a complete geometric model.However,the existing point cloud registration algorithm which is used for complex point cloud obtained by objects with complex structure curved surface is lack of pertinence,so the accuracy of registration is not ideal.Therefore,this paper studies registration algorithms of complex point set.The main research is as follows:1)For large-scale and complex point cloud set,we propose a registration method based on adaptive Gaussian mixture model.The point cloud data is represented as adaptive Gaussian mixture model,and the similarity probability is taken as the objective function.The Expectation Maximization Algorithm and Fast Gauss Transform are used to estimate the parameters.The experimental results show that the proposed algorithm not only improves the accuracy of point set registration,but also improves algorithmic time complexity.2)For small scale and complex point cloud set,we divide point set registration into two stages: rough registration and accurate registration.In the rough point set registration stage,the neighborhood curvature feature and Normalized Zero-mean Cross-correlation Coefficient are used to construct corresponding point.Then the matrix transformation is carried out by using the quaternion method.In the accurate registration stage,we propose the iterative nearest Closest algorithm based on the mixed distance.The algorithm uses the mixed distance function as the objective function.Then we use Artificial Bee Colony algorithm to search for the nearest point.The experimental results show that the algorithm not only improves the accuracy of registration,but also reduces the cost of computational.3)In this paper,a point cloud registration system is implemented using the above algorithms.The system can preprocess the point cloud data,and effectively complete the registration operation of the complex point cloud data,and output the registration results.