In recent years, artworks have received extensive attention due to their unique expressions and conveyed ideas, and interactive artworks have also aroused the interest of the public and researchers. At the same time, with the rapid development of computer science and technology, emerging technologies such as gesture recognition have brought new development directions to interactive artworks. This paper focuses on the application of gesture recognition technology in interactive artworks. By designing a deep learning structure to estimate the 3D hand pose from regular RGB images, the gesture recognition accuracy is maintained while reducing the amount of network parameters, and the gesture recognition network is lightened so that it can be applied to interactive artworks.