Research on visible and infrared image fusion method based on transformer

In autonomous driving, diverse and complex environments pose significant challenges for all-weather perception, especially under poor lighting, rain, fog, or snow. Relying solely on visible cameras often fails to capture sufficient information, limiting system performance. This paper proposes a Transformer-based visible-infrared image fusion method. It combines convolutional layers and Swin Transformers with residual connections for local and global feature extraction. A cross-modal fusion module is designed for deep interaction of complementary information. CNN and Transformer blocks are used for feature reconstruction. Experimental results demonstrate the proposed method significantly improves fusion quality, offering an effective solution for autonomous driving in challenging environments.