Driveable 3D Human Reconstruction Focusing on Facial Precision Enhancement

Three-dimensional (3D) digital human reconstruction and driving technologies are increasingly used. The most commonly used method in the field of 3D digital human body reconstruction and driving is based on parameterized models, such as SMPL, which uses parameters to represent the shape of the human body and bone posture, and uses skin algorithms to calculate the skin. However, these algorithms typically do not specifically design and optimize the face region, so reconstruction and driving can cause the loss of facial identity information. This paper proposes a method that combines 3D human body reconstruction and driving methods with 3D facial precision enhancement reconstruction methods. It first reconstructs the human body and face separately from single-view RGB images or videos, and then fuses the human body and face models into a 3D digital human with facial precision enhancement, enabling it to be driven by the actions and expressions of characters in external videos, while maintaining facial identity information unchanged.