The network video training mode of five step fist has become the mainstream training mode at present, because it is not limited by time and region. However, there are also some problems, such as irregular movements of practitioners, due to the influence of subjective human eyes, which will damage their health. Therefore, this paper proposes a pose recognition and correction method of the five step fist based on OpenPose. By identifying the standard movements and the movements of the practitioners, the key points are extracted, the deviation and similarity between the practitioners and the standards are calculated, and the corresponding correct and effective correction suggestions are given. These works can provide strong technical support for the training and cultural dissemination of the five step fist.