Exploiting Visual Quasi-Periodicity for Automated Chewing Event Detection Using Active Appearance Models and Support Vector Mach