Master's Thesis Award Winners 2023

Development and Evaluation of an Algorithm for Line Segmentation from Point Clouds for Fiber Composite Systems

Ronja Miehling
Supervisor: Prof. Dr.-lng. habil. Dr. h.c. Volker Schwieger
Advisor: Laura Balange

Determining the geometry of structural components is an essential part of any construction process, as deviations can lead to a change in the structural design under certain conditions. In this work, building objects made of fiber composite systems are considered. In order to determine the very complex geometries of these objects, the components are scanned by a laser scanner (Figure 1 ). An algorithm is designed for the resulting point clouds, which detects (segments) the individual fiber lines as weil as their intersection points. For the line segmentation, different methodologies of image and point cloud processing are used. The methods investigated include iterative Hough transform, orthogonal least squares adjustment, and RANSAC. For the next segmentation step for the point clouds, voxels and k-means clustering are used (Figure 2). The results of the algorithm are the coordinates of the intersection points (Figure 1, right), the parameters of the segmented lines as weil as the points of the entire point cloud belonging to the line (Figure 3). Since an evaluation of the results is difficult due to missing ground truth data, additional point clouds were simu.lated to know ground truth. In general, the results show that lines and their intersections are correctly detected to a high degree (Figure 4). However, challenging cases e.g., low point densi_ty led to the fact that fibers cannot be recognized. In addition, duplications of lines or intersections can still occur in these challenging cases.