3D Point Cloud Plane Segmentation Method Based on RANSAC And Support Vector Machine

Recently, three-dimensional (3D) laser scanning technology has gradually become a main method of retrieving geometric information of objects and scenes.By processing the point cloud data obtained,we can implement 3D object recognition and the automatic reconstruction of indoor and urban environments,which are significant contents of the research fields of computer vision and robotics.As one of the primary tasks of point cloud processing, plane segmentation has also drawn attention of scholars from all around the world and become a very promising research area. Among different plane-segmentation methods,Random Sample Consensus (RANSAC) is a highly robust method and enjoys a strong capability of anti-interference.However,it suffers from the problems of generating spurious planes and relatively low accuracy in plane segmentation of complex environments.Two improved methods based on basic RANSAC are proposed in this study to enhance the accuracy of plane segmentation.Our method use Support Vector Ma-chine(SVM),the supervised learning model used for classification of the point clouds,into basic RANSAC to predict the category of a certain point according to its three-dimensional coordinate and rgb value.Every category corresponds to a plane.In this way,the high accuracy of SVM in classification and the preciseness of taking every point into consideration can result in the improved accuracy of plane segmentation.The results of experiments validate the effect of our method.