Since the emergence of computer vision, pedestrian detection technology has been widely concerned by scholars at home and abroad. The technology is widely used in real-time monitoring, intelligent transportation and human-computer interaction. Most of the existing pedestrian detection systems have problems such as large program volume, high dependence on operating environment, and inability to guarantee recognition rate in complex situations. In order to improve these problems, a pedestrian detection system based on OpenCV vision library is designed, which uses deep learning framework for reasoning and separates model training from prediction. Firstly, YOlOv5 based on PyTorch framework is rewritten under OpenCV DNN framework, and the core algorithm of target recognition including intersection over union, non-maximum suppression and other modules is written. Then the user interface of pedestrian monitoring system is developed on Windows platform. Finally, multi-scene experiments are carried out to verify that the designed pedestrian monitoring system has high recognition accuracy.