Aiming at the problem of low accuracy of multi-scale pedestrian target detection in low-illumination environment, a pedestrian detection method based on YOLOv5s and infrared and visible light image fusion is proposed.Firstly, the Generative Adversarial Network (GAN) is used to make a data set of the fused visible light and infrared images, and the influence of external factors on pedestrian detection can be reduced by combining the advantages of two kinds of images, and then, the SENet channel attention module is introduced into YOLOv5s to make the network pay more attention to the highlighted target, so as to improve the accuracy of pedestrian detection.Finally, the structure of YOLOv5s network is optimized, part of the convolution layers are deleted, and the activation function is modified to maintain the high real-time performance of the algorithm.Experimental results show that a detection model with higher mAP can be obtained by using the fusion image data set for training than using the visible light data set or the infrared data set.The improved SE-YOLOv5s algorithm effectively improves the mAP of pedestrian detection while maintaining the high real-time performance of the original algorithm.