Advancement in the head pose estimation via depth-based face spotting

Abstract

Head pose estimation is not only a crucial preprocessing task in applications such as facial expression and face recognition, but also the core task for many others, e.g. gaze; driver focus of attention; head gesture recognitions. In real scenarios, the fine location and scale of a processed face patch should be consistently and automatically obtained. To this end, we propose a depth-based face spotting technique in which the face is cropped with respect to its depth data, and is modeled by its appearance features. By employing this technique, the localization rate was gained. additionally, by building a head pose estimator on top of it, we achieved more accurate pose estimates and better generalization capability. To estimate the head pose, we exploit Support Vector (SV) regressors to map Histogram of oriented Gradient (HoG) features extracted from the spotted face patches in both depth and RGB images to the head rotation angles. The developed pose estimator compared favorably to state-of-the-art approaches on two challenging DRGB databases.