A face tracking method using dominant orientation templates and pose estimation

This paper proposed a method for face tracking and pose estimating by combining two algorithms, namely, the dominant orientation templates (DOT) [1] and the pose from orthography and scaling with iterations (POSIT) [2]. The concept of DOT is measuring the similarity between an input image and a reference image by matching from gradient orientations between a set of dominant image orientations and a set of dominant template orientations. The DOT algorithm is fast and accuracy as studied in [1]. The POSIT algorithm estimates 3D pose of an object and finds the rotation matrix and the translation vector of the object. To apply the DOT algorithm to face tracking, we selected four important face features which are left eye, right eye, nose, and mouth from different postures of a face. The gradient orientations are, then, extracted from these face features and these gradient orientations are combined to create the gradient orientation templates. We use these templates for locating the four face features on the face image. The positions of all face features are input to the POSIT algorithm where the face rotation matrix and transition vector in 3-D is estimate. The experimental result show the propose method can get satisfactory result.