RDSHNet

Infrared imaging exhibits strong environmental robustness in low-visibility scenarios such as nighttime and rainy/foggy conditions; however, due to low signal-to-noise ratio, blurred boundaries, scarce textures, and pronounced background clutter, real-time vehicle detection still faces dual challenges: weakened structural details and unreliable high-level semantic representations. To achieve a better trade-off between accuracy and efficiency, this paper proposes RDSH-Net, a Real-time Detail-Semantic Hybrid Network for real-time infrared vehicle detection.