Real-Time Head Pose Estimation Using Random Regression Forests

Automatic head pose estimation is useful in human computer interaction and biometric recognition. However, it is a very challenging problem. To achieve robust for head pose estimation, a novel method based on depth images is proposed in this paper. The bilateral symmetry of face is utilized to design a discriminative integral slice feature, which is presented as a 3D vector from the geometric center of a slice to nose tip. Random regression forests are employed to map discriminative integral slice features to continuous head poses, given the advantage that they can maintain accuracy when a large proportion of the data is missing. Experimental results on the ETH database demonstrate that the proposed method is more accurate than state-of-the-art methods for head pose estimation.