Facial expressions are complex to model since they involve various factors, mainly psychological, biomechanical and sensory. While facial performance capture provides good facial expressions, it is costly and difficult to use for real time interaction. A number of tools and techniques exist to automate the facial animation related to speech or emotion, but there are no tools available to automate the facial expression related to physical activity. This leads to unrealistic characters, especially when a 3D character performs intense physical activity. The purpose of this research is to highlight the link between physical activity and facial expression and propose a data-driven approach to realistically simulate the facial expression while leaving creative control. First, motion capture was used to gather information relating biological, mechanical and facial expressions data. This involved two different motion capture sessions, each gathering specific information and each involving several participants. This information was used to train machine learning models that can predict facial expressions from inputs such as the 3D character motion, the weights lifted, etc. The proposed approach can be used with realtime, pre-recorded or key-framed animations making it suitable for video games and movies as well.
| Date | 6 Jul 2012 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Eric Paquette (Supervisor) |
|---|
Boukhalfi, T. (Author),
Paquette (Supervisor),
6 Jul 2012Student thesis: Master's thesis › Master in Engineering: Engineering