A Comprehensive Review of Existing Datasets for Off-road Autonomous Vehicles

• Main Authors: Szabo, Lorant, Weltsch, Zoltan
• Format: Conference Proceeding
• Language: English
• Subjects: Annotations; Autonomous vehicles; Convolutional neural networks; Datasets; Navigation; Off-Road; Semantic Segmentation; Synthetic data; Training; Virtual reality
• ISSN: 2767-9438
• DOI: 10.1109/SAMI60510.2024.10432820

Developments concerning autonomous vehicles in the public consciousness primarily represent function development in urban and structured environments for the majority of both ordinary people, researchers and engineers, mainly due to the availability of large amounts of high-quality databases. However, function development for off-road conditions presents numerous challenges, from navigating difficult terrains to understanding varied environmental contexts. An overview of the currently available datasets supporting off-road autonomous driving is provided here, primarily from the perspective of developing environmental perception as a capability. This study highlights the need for more comprehensive and diverse datasets, presenting the limitations of currently achieved datasets. It makes suggestions for future directions in the creation of datasets, emphasizing the importance of synthetic datasets and their relationship compared to those collected in reality. The insights highlighted in this document can serve as a valuable resource for researchers and developers dealing with autonomous vehicles, especially for those primarily developing environmental sensing functions, and within that, semantic segmentation.