Research on Facial Expression Animation Technology Based on OpenFACS

This paper focuses on three core research objectives: (1) Optimize the AU detection algorithm based on OpenFACS to improve the accuracy of identifying subtle facial movements; (2) Establish a real-time mapping framework between OpenFACS AUs and 3D facial blend shapes; (3) Validate the proposed technology through objective performance metrics and subjective user evaluations. The research methods include: (1) Using the CK+ (Cohn-Kanade) and BU-3DFE (Binghamton University 3D Facial Expression) datasets for model training and testing, combined with preprocessing steps such as facial landmark detection and image normalization; (2) Designing a hybrid CNN-LSTM (Convolutional Neural Network-Long Short-Term Memory) model to fuse spatial features (from CNN) and temporal features (from LSTM) for AU detection, addressing the dynamic nature of facial expressions; (3) Developing an AU-blend shape weight matrix based on geometric analysis of facial muscles, and optimizing real-time rendering using GPU acceleration. This research enriches the application of OpenFACS in facial expression animation, improves the accuracy and real-time performance of the technology, and provides a low-cost, standardized solution for developers in VR, games, and HCI. Future work will focus on integrating multimodal data (e.g., eye gaze and voice) to enhance expression authenticity and expanding the framework to support dynamic lighting and pose variations.