A multimodal recognition and detection method for road scenes based on the improved YOLOv11

In the current process of road visual detection, traditional object detection models perform poorly in complex environments such as low light and nighttime conditions. Particularly in the case of multimodal data object detection involving visible light and infrared images, existing models lack adaptability and robustness to different modalities. This paper proposes an HI-YOLOv11 object detection method. It combines the Histogram Transformer Block (HTB) module to enhance the image information expression ability between modalities in nighttime and infrared light conditions. At the same time, it uses the Inner-IoU Loss function to finely optimize the target position. Experiments show that the model in this paper has achieved mAP (50) and Recall rates of 95.76% and 91.84% respectively, which are 1.85% and 2.18% higher than the original model. Therefore, the method proposed in this paper has better robustness in detection under multi-lighting and complex environments.