In complex environments, pedestrian detection faces problems such as low visibility, poor image contrast and noise interference, which seriously affect the accuracy and safety of detection. In this paper, a pedestrian detection method based on improved dual-stream convolution channel based on visible and infrared image fusion is proposed. Firstly, taking advantage of the thermal infrared image under low light conditions, a dual-stream feature extraction network was constructed to enhance the fusion effect of visible light and infrared features. Then, the lightweight YOLOv8n is used as the basic network, and the features are extracted through the parallel improved CSPDarknet53 to ensure real-time performance in complex scenes. Finally, we conduct experiments on the LLVIP dataset, and the results show that the proposed method achieves 0.984 in average accuracy (mAP@50), which is 4.9% higher than that of the basic network, and significantly reduces the number of parameters by 20.1% and the amount of computation by 17.8%. This method has superior performance in the accuracy, robustness and real-time performance of image detection, and provides a new solution for pedestrian detection.