Enhancing Pedestrian Recognition at Night: A Method Based on Global-Texture Fusion (GTF) Network
Yu Yu, Xiyu Liu, Yifu Guo, Yujia Liu · IEEE Sensors Journal · 2025
Infrared–visible image fusion plays a crucial role in enhancing night-time visibility and addressing challenges posed by low lighting, occlusion, and other adverse conditions, thereby benefiting a range of applications such as night vision, environmental monitoring, autonomous driving, and UAV-based surveillance. However, existing transformer-based methods demand substantial computational resources, whereas traditional CNN-based methods often struggle to capture sufficient global information. In this work, we propose a Global-Texture Fusion (GTF) Network, an enhanced CNN-based framework that integrates a Global Block to extract large-scale context and a Texture Block to preserve fine-grained details. This design reduces computational overhead while achieving superior fusion performance. Experimental results on three public benchmark datasets, RoadScene,TNO and LLVIP demonstrate that GTF outperforms state-of-the-art approaches in terms of entropy, mutual information, PSNR, and structural similarity, showcasing clear improvements in both global scene understanding and local texture preservation. Our code will be made publicly available at: https://github.com/QLYYLQ/GTF.