University of Twente Student Theses

Login
As of Friday, 8 August 2025, the current Student Theses repository is no longer available for thesis uploads. A new Student Theses repository will be available starting Friday, 15 August 2025.

Feature-level fusion of 2D images and 3D LiDAR point clouds for semantic segmentation

Nikolov, Andrey (2025) Feature-level fusion of 2D images and 3D LiDAR point clouds for semantic segmentation.

This is the latest version of this item.

[img] PDF
4MB
Abstract:Semantic segmentation is a crucial task in autonomous systems, including those used in driving, robot navigation, and medical diagnosis. While there are methods for 2D segmentation using convolutional neural networks (CNN) and 3D segmentation using 3D models, the complementary nature of 2D data and 3D data should not be ignored. This research investigates multimodal fusion of 2D images and 3D LiDAR point clouds for semantic segmentation in structured and unstructured environments. Building on the DeepViewAgg framework, we aim to investigate the impact of feature fusion on semantic segmentation compared to 2D- and 3D-only models. The methodology involves training a model for each modality and evaluating its performance. On KITTI-360, fusion improves mean IoU from 54.20 (3Donly) and 56.70 (2D-only) to 57.53, with the largest gain on thin classes such as ’pole’ (+21.3 points). In the WildScenes natural dataset, it achieves 33.0 mIoU, outperforming 2D and 3D baselines with a margin of 5.0 points. These trends demonstrate that multimodal fusion can outperform single modalities, particularly in scene elements with complementary 2D-3D cues.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Business & IT BSc (56066)
Link to this item:https://purl.utwente.nl/essays/107516
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page