Extraction Of 3D Line Segments From MLS Point Clouds

Straight-line segments in light detection and ranging(LiDAR)point clouds play an important role in many applications.Benefiting from the highly developed LiDAR devices,using large-scale point clouds in many tasks,such as object detection and classification from 3D point clouds,have become a hot topic.However,only a few studies focused on low-level feature extraction,for example,3D line detection.Different from detecting lines from 2D images,which has been fully investigated and many related algorithms have been proposed,in the field of 3D line segments detection from LiDAR point clouds,there is still room for improvement.In this master thesis,we present a novel and efficient method to extract 3D line segments from LiDAR point clouds directly.Unlike other methods which either extend the application of 2D line segment detection in 2D image to 3D space or first detect the surfaces and then extract sur-face boundaries as line segments,we proposed a point-based method with a bottom-up strategy.First,a geometric feature,angular gap,based on the local neighborhood,is introduced to detect the edge candidate point.Then,for each edge candidate,an edge candidate orientation is calcu-lated.Those edge candidates are segmented into original lines via region growing and RANSAC line fitting.Finally,a line optimization procedure is proposed to eliminate outliers and merge adjacent 3D line segments.Moreover,we introduce a new accuracy assessment evaluating the detection of 3D line segments,involving the concepts of correct extraction,over-segmentation,under-segmentation,and noise,which are quality indicators of the extracted 3D line segments.The proposed method is tested on various types of LiDAR point clouds acquired in different ways.We also compared the method performance with other state-of-the-art method and pro-vided the visualization and quantitative evaluation comparison results.The experimental results demonstrate our method is efficient and effective in 3D line segments extraction while delivering more accurate and complete line segments than the comparative approach.The proposed line segment evaluation metrics show effectiveness in indicating the extraction quality and method performance.