Research On Some Key Technoligies Of3D Point Cloud Data Processing

With the rapid developments of computer and3D measurement technology,3D pointcloud data has been used to many fields more and more widely, such as reverseengineering，industrial inspection, protection of historical relics and so on. The originalpoint cloud data is often faced many problems which are caused by the devisemeasurement error, measurement range, surroundings and etc. So we must deal with themby outlier auto-detection, smoothing, auto-registratin and etc. The content and main workof this paper as follows:1. An outlier auto-detection algorithm is implemented which is based on the averageneighborhood algorithm. It uses neighborhood distance deviation as the judgement objectto reduce its sensitivity to parameter settings greatly, and improve s its stability. Theexperiment results show that the method can detect the discrete and next to modelboundaries outliers easily, and maintain the borders. The effect of the method is good.2. A3D data smoothing algorithm based on moving least squares is implementedwhich uses the advantage of moving least squares on fitting surface. The key of thealgorithm is the selection of appropriate basis function and weight function. Theexperiment result shows that the method has a good smoothing effect for the data with alot of noise, and maintains the characteristics of the model to the greatest extent.3. A point cloud data auto-registratin technology is implemented. In the process ofcoarse registration, a correcting mismatching marked points method based on RANSACwas proposed after finding the original matching marked points. The method divides allmatching marked points into inner points and outer points according to the selected targetmodel and related criteria, calculates the current optimum target model parameters usingthe inner points and finally calculates the best parameters after a certain times of randomsampling. It effectively removes the mismatching marked points which are generated inthe process of point cloud auto-registration. Analog experiment and registration examplesdemonstrate that the method is practicable and improves the stability of point cloud auto-registration effectively.The precise registration uses iterative closest point algorithmwhich is used widely, and the algorithm is divided into5steps: determination of theinitial set of corresponding points by computing closest points, removal of the errorcorresponding points, computation of the coordinate transformation matrix, coordinatetransformation of the point-clouds and interative. The experiment result shows that thealgorithm can optimize globally and improve the accuracy of point cloudauto-registration.