A Noise Reduction Method For Scanned 3D Point Cloud Based On Density Clustering And Majority Voting

Three dimensional scanning technology is widely used in reverse engineering,cultural relics protection,industrial production,criminal investigation,three-dimensional fax and other fields.Point cloud data refers to the data recorded in the form of points obtained by scanning and each point corresponds to a three-dimensional coordinates.In the process of 3D scanning,the error of the instrument itself or the external factors in the process of scanning will be introduced into the abnormal points.If it is not timely removal of abnormal points,it will result in a larger experimental error.In this paper,the problem of noise data in the point cloud processing method of the three-dimensional scanning technology is studied as the research direction.There are two types of abnormal points in the scanning data.The part of the abnormal points are far from the overall point cloud data,which are called discrete abnormal points.And the other part are mixed in the point cloud data,which are called non-discrete abnormal points.For the detection of discrete abnormal points is relatively easy,but for the non-discrete abnormal points,it needs more determined conditions to increase the calculation and the complexity of the algorithm.At present,the processing algorithm has the disadvantages of complex processing flow,long processing time and poor applicability.The method proposed in this paper is a two stage abnormal point processing method.The first stage points to discrete abnormal points,which is mainly determined by the degree of data points distribution.Then the data clusters are divided into normal clusters,suspected clusters and the abnormal clusters.The second stage is to face the non-discrete abnormal points and the criterion is to determine the degree of closeness with the fitting surface.In the first stage,the data points are classified into a number of data clusters.After calculating the data density of each cluster,the data clusters are divided into normal clusters,suspected clusters as well as the abnormal clusters.The second stage is the use of normal clusters in the collection to make neighborhood voting judgment of each point in the suspected clusters.According to a suspected point to the fitting surface distance and the principle of majority decision,suspected point will be decided whether it is an abnormal point.Eventually,it removes all abnormal points and preserve all normal points to get a reasonable three-dimensional point cloud data model.In order to facilitate viewing processing results in each stage of noise reduction algorithm,this paper uses Eclipse and OPENGL to develop a data noise reduction procedures,including point cloud acquisition,processing,display and storage modules.It achieves data import,data processing,image display and data export.Sets of real scan data are imported to data noise reduction system.The images display that the noise reduction method can effectively remove the abnormal points of 3D point cloud data model in manufacturing the working modules and can maintain the good characteristics of the model surface,accelerating the processing efficiency.