RGB and Thermal Image based Human Intrusion Detection in Off-Road Scenarios: A New Dataset and Benchmark

View at publisher Summarise with VIRAT
Authors: Reshu Yadav, Amit Mitra, Rahul Kala
Cited by: 0
References: 20

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

We introduce a novel RGB and thermal dataset specifically designed for human intrusion detection in challenging off-road environments, captured using multiple synchronised RGB cameras and a thermal camera mounted on autonomous ground vehicles (AGVs). The dataset encompasses diverse terrains, varying lighting conditions, and dynamic human activities such as walking, running, and crawling. To assess its effectiveness, we benchmark several state-of-the-art object detection algorithms on this dataset, evaluating their robustness and limitations in complex, unstructured off-road scenarios. Our results provide critical insights into the capabilities of image-only perception for reliable human detection, particularly under low-light and cluttered backgrounds, which are crucial for safe AGV operation. Experimental results show that DINOv2 gives a very good performance on both RGB and Thermal images, while the performance on RGB images is generally better. The performance of transformer-based models, including Grounding DINO, DETR and RT-DETR, along with YOLOv8, is better compared to more classic deep-learning approaches. The dataset is available at https://github.com/RoitYaadav/DT

Fields of Study & Subjects

Advanced Neural Network Applications Autonomous Vehicle Technology and Safety Video Surveillance and Tracking Methods

Publication details

Journal / source
IEEE International Conference on Industrial and Information Systems (ICIIS)
Pages
370–375
Publication date
16 January 2026
Language
EN