Multi-Source Aggregation for Automatic 3D Point Cloud Annotation in Road Scenarios
DOI:
https://doi.org/10.24425/ijet.2026.6312Abstract
The creation of high-quality 3D annotations for lidar point clouds involves human experts, it is a time-consuming and costly process, limiting the development of large datasets for autonomous driving. This work reviews existing datasets, annotation tools, and automated and semi-automated methods for generating 3D annotations in road scenarios, with particular emphasis on approaches based on point clouds, multimodal data, and weakly supervised learning. Based on this review, an annotation-support method is proposed that combines 3D detections generated by multiple independent sources. Their outputs were aggregated using an IoU-based strategy, and the resulting annotations were evaluated using Average Precision (AP), Intersection over Union (IoU), and detection statistics. The experiments show that aggregating detections can improve annotation quality when only a small amount of training data is available. As the amount of training data increased, the benefit of aggregation decreased, reaching no improvement for the full training set. These results indicate that combining heterogeneous annotation sources can be an effective way of supporting the creation of 3D point cloud annotations, particularly in the early stages of dataset development.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 International Journal of Electronics and Telecommunications

This work is licensed under a Creative Commons Attribution 4.0 International License.