Research On Laser Point Clouds Repair Based On Low Rank Decomposition

In recent years,with the continuous breakthroughs of 3D scanning technology,the point cloud model has become an emerging digital media after sound,image and video.The point cloud is widely used in reverse engineering,computer science,architectural design,cultural relic protection and film,games and other areas of technology and entertainment.Different from traditional digital media,the point cloud has strong realism and plentiful details,and is gradually becoming an important data source for various industries,promoting the development of researches in various industries.For the point cloud,the integrity and accuracy of its data are the basic conditions for its subsequent extensive applications.However,in the actual scanning process,especially when scanning with a laser,due to the geometric and optical characteristics of the object itself and the mechanical stability of the scanning system,the actual point cloud often has a large number of holes and burrs on its surface.The existence of these missing data will have a serious impact on the subsequent model reconstructions,so that the complete and effective surface morphological features of the measured object cannot be obtained in such conditions.Therefore,the repair of missing data and the elimination of point cloud noise have become the key to breaking through the research and application bottleneck of point cloud.To solve the problems above,this paper studies the optical working principle and mechanical structure characteristics of the laser point cloud scanning system,and analyzes the causes and characteristics of the distortion,such as holes,burrs and jitters of the point model generated.On this basis,point cloud data collection and analysis are performed on different objects to be tested,and the accuracy and reliability of the above theoretical analysis are verified by experiments.Then,this paper compares the principle,advantages and disadvantages of the traditional isolated point detection method and combines the statistical-based method with the distance-based method to design a set of efficient and accurate isolated point detection algorithm based on K nearest neighbor method and statistical screening method.The effectiveness and accuracy of the proposed method are proved by comparative experiments.Finally,this paper applies the idea of low rank decomposition to the point cloud.To optimizes the accuracy of the low rank decomposition method,this paper constrains the classical low rank decomposition model according to the point cloud features analyzed in the previous sections.Experiments show that the proposed method can effectively repair the point cloud model,which is much better than the traditional low rank decomposition method.