Contribution

Comparison of neural networks road detection in off-road environments

---

Abstract

As unmanned ground vehicles (UGVs) are more frequently deployed in unstructured environments, there is a growing need for robust road and terrain detection systems. The ability to navigate autonomously in challenging Terrains depends on the effectiveness of computer vision models. Off-road environments encompass rugged terrain, forest roads, agricultural fields, and more, characterized by dynamic changes and unpredictable obstacles. UGVs must discern drivable ground to enable effective navigation while identifying and circumventing obstacles in real-time. This paper investigates different sensor-based and neural network-driven approaches to address these challenges, focusing on the critical task of identifying forest roads in off-road environments. Using different sensors, we assess their effectiveness in different environmental conditions through a comprehensive comparative analysis of three neural network architectures. Our results highlight the strengths and limitations of different sensor modalities and neural network models. They provide insight into their performance under adverse conditions such as overexposed images, complex shadows, and dense vegetation on forest roads. This research provides valuable insights into developing robust off-road navigation Systems essential for advancing autonomous ground vehicle Technology.

How to Cite:

Oberpertinger, J., Eder, M. & Steinbauer-Wagner, G., (2025) “Comparison of neural networks road detection in off-road environments”, ARW Proceedings 25(1), 109-114. doi: https://doi.org/10.34749/3061-0710.2025.18