In helmet detection, issues include excessive network resource usage, high time delay, and inability to detect in real time when transmitting surveillance videos. A helmet-wearing detection system leveraging cloud edge collaboration technology was developed. The improved YOOv5 helmet detection model is trained to deal with a large amount of data and complex tasks through powerful computing resources in the cloud, and then the lightweight model is migrated to the terminal monitoring equipment using edge computing technology for reasoning and application. While the terminal monitoring equipment completes real-time data collection, the video stream data does not need to be connected to the cloud service software, Directly through the built-in edge computing platform, the real-time automatic perception and local alarm of violations are realized, which greatly saves bandwidth, reduces latency, ensures data security, and reduces the workload of on-site management personnel. By further establishing a smart IoT security control platform and implementing information linkage through cloud edge collaboration technology, the real-time and interactivity of the system have been greatly improved. The experimental results demonstrate that the system achieves an average accuracy of 97.8% on the SHWD dataset and the average processing delay of the video is 0.9 s. It can quickly and accurately identify whether the operator is wearing a safety helmet.