Multi-view Constrained Local Models for Large Head Angle Facial Tracking

We propose Multi-View Constrained Local Models - a simple but effective technique for improving facial point detection under large head angles, such as in a car driving setting. Our approach combines a global shape model with separate sets of response maps targeted at different head angles, indexed on the shape model parameters. We explore shape-space division strategies and show that, as well as outperforming the traditional method, our approach also provides a marked speed-up which demonstrates the suitability of this technique for real-time face tracking.