Abstract: Due to the ubiquity of RGB cameras in mobile computing devices such as virtual reality headsets, the 3D hand pose estimation technology based on a RGB image has broad application prospects and research value, which becomes a research hotspot in the field of computer vision in recent years. Thanks to the development of deep learning technology, algorithms related to 3D hand pose estimation emerge in endlessly. This paper reviews and summarizes the 3D hand pose estimation technology. Firstly the relevant work on 3D hand pose estimation is briefly described, and the current challenges it faces are pointed out; then the algorithms of 3D hand pose estimation from a single RGB image are reviewed, and the existing model-based methods and model-free methods are discussed; then the relevant datasets and evaluation criteria are summarized; finally the development prospects of this technology are discussed.