Bionic Hand System for 3D Hand Reconstruction Using a Single View

This paper proposes a simulation system for 3D hand reconstruction using a single view, which is dedicated to play a role in efficiently reconstructing 3D hands and accurately controlling bionic robotic hands in the fields of constructing 3D hands and bionic robotic hand control. The 3D hand reconstruction method by using SimpleHand network structure and applying it to bionic hand control and simulation. Using the FreiHAND dataset, the 3D coordinates of 21 joint points of the hand are obtained by SimpleHand network inference and rendered by MANO model to evaluate the network performance, which is also combined with several evaluation metrics. Subsequently, these 3D joint point coordinates were converted into joint parameters of the E3M5 model of the Mujoco bionic hand to achieve accurate movement control of the bionic hand. In order to improve the reconstruction accuracy and training efficiency of the model, this paper uses Bayesian Optimization to tune the key hyperparameters such as learning rate, weight decay and batch size. The experimental results show that Bayesian Optimization improves the reconstruction accuracy of the model, and the F@05 metric is improved by 2.2%. This paper demonstrates the effectiveness of SimpleHand network in 3D hand reconstruction task and verifies its potential for application in the field of gesture recognition and robot control.