University of Twente Student Theses


Point cloud classification and segmentation of catenary systems

Vieth, Z.J. (2022) Point cloud classification and segmentation of catenary systems.

[img] PDF
Abstract:The Dutch railway is transitioning towards digitization to improve the efficiency of maintenance and reduce material waste and system downtimes. Strukton Rail and the Saxion University of Applied Sciences are working on deep-learning-based methods to digitize the Dutch railway's catenary system. This paper explores the development of a data pipeline for the reconstruction of 3D models from point cloud-based railway scenes utilizing a provided CAD catalogue. We used a two step approach to achieve this goal: segmentation and object retrieval from the catalogue. In the first step, the deep learning-based segmentation method PointNet++ was used. We used a RANSAC and point pair features-based template matching implementation in the second step. Testing with three substantially different catenary components generates final region overlap scores ranging from 60% to 100%, depending on the quality of segmentation results, down-sampling parameters, and the number of RANSAC iterations. These results confirm the applicability of the approach. The manual steps of the current pipeline suggest a future need for the use of instance segmentation models, component-variant consideration, and automation of CAD model placement.
Item Type:Essay (Bachelor)
Saxion University of Applied Sciences, Enschede, Netherlands
Strukton Rail, Utrecht, Netherlands
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Creative Technology BSc (50447)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page