Design And Implementation Of Point Cloud Registration System Based On Improved ICP Algorithm

As an important direction of computer vision and pattern recognition,point cloud registration is widely used in reverse engineering,cultural relic restoration and virtual reality etc.At present,the robustness of point cloud rigid body registration method to outliers and measurement noise needs to be further improved.Besides,the traditional iterative closest point(ICP)algorithm has been well applied in point cloud rigid registration,but how to deal with complex and changeable point cloud affine registration is still an urgent problem.Therefore,it is of great theoretical and practical significance for accelerating the development of computer vision,pattern recognition,improve the development of production automation and other fields to deeply explore effective point cloud registration methods and improve and optimize point cloud registration technologies.Based on the in-depth analysis of ICP algorithm,this thesis proposes improved methods for estimation problems of both rigid registration and affine registration of point clouds.And the corresponding point cloud registration system is designed and implemented as well.The main contents of this thesis include the following three aspects:(1)By using the RGB-D data obtained by Kinect depth camera,a point cloud rigid registration method is presented in this thesis to solve the problem of the estimation of point cloud rigid transformation relationship.In order to improve the accuracy of point-pair matching,the corresponding relationship of point clouds is established with the help of color information.Considering that the pseudo Huber loss function is not sensitive to outliers and noise points,the pseudo Huber loss function is introduced to optimize the rigid registration model.The experimental results show that,the proposed method can effectively suppress the influence of outliers and noise on the rigid registration results compared with the traditional algorithm.(2)Aiming at the problem of affine registration estimation of point clouds,an affine registration method of point clouds is presented in this thesis.In order to improve the consistency of color information,the hue value information of point clouds is added as a constraint condition to establish a good correspondence between point clouds.And the pseudo Huber loss function is adopted to establish the optimized objective function for registration.In order to avoid the solution of the method falling into the local minimum,the alternative iterative strategy of Levenberg-Marquardt(LM)algorithm and singular value decomposition(SVD)is adopted to optimize.The experimental results show that,the proposed method can solve the affine registration problem of point clouds accurately and robustly,and reduce the influence of outliers and noise compared with the traditional algorithm.(3)A point cloud registration system is designed and implemented in this thesis.The system encapsulates the traditional algorithms and the proposed methods into the method library.The system encapsulates the traditional algorithm and the method proposed in this thesis into the method library to help users quickly complete the point cloud registration and evaluate the performance of each method.And this thesis discusses the requirements analysis,overall and module design,implementation and testing of the system.The practical application of point cloud registration is realized by using this system.