Aiming at the problems of high missed detection rate and low detection accuracy in current urban street scene pedestrian detection methods, an improved YOLOv7 urban street scene pedestrian detection algorithm was proposed. This method integrates the CA attention mechanism and the SE network to design the CA-SENet attention mechanism to enhance the network's attention to pedestrians and improve the model's detection accuracy for pedestrians. An improved atrous spatial pyramid pooling module is introduced and embedded in the connection between the backbone network and the feature enhancement network to capture multi-scale features of the image and reduce the missed detection rate of the model. The Wise-IoU loss function is used to replace the original CIoU to optimize the quality evaluation of the anchor box. After experimenting with the algorithm on public datasets and self-made pedestrian datasets, the results show that the improved YOLOv7 model can detect pedestrian targets more effectively.