Wearing a helmet can protect the heads of production workers from falling objects. At present, there are many studies on safety helmet detection, but there are relatively few studies on the detection of the complex posture of workers, and it is not possible to determine whether the helmet-wearing position can protect the heads of workers. At the same time, there are problems in the rolling workshop such as a large space span, a wide range of operating equipment, a chaotic environment, a large difference between day and night light, dazzling light, and a large variation range of monitoring target scales. It increases the difficulty of helmet-wearing detection. To solve the above problems, a detection scheme for helmet-wearing in a steel rolling workshop was designed. Firstly, the improved YOLOv7 model was used for helmet and pedestrian detection, then the improved YOLOpose model was used for attitude estimation to detect the position of workers’ heads, and finally, the Hungarian algorithm was used to match the position of helmets and heads to determine whether workers wore helmets correctly.