Unconstrained Head Pose Estimation with Constrained Local Model and Memory Based Particle Filter by 3D Point Distribution Models

In this paper, a novel and efficient method was proposed for unconstrained head pose estimation of the human face and robustness to changes of pose, position and facial expression. A Constrained Local Model (CLM) by a 3D Point Distribution Model (PDM) was proposed for locating 2D facial landmarks in optional poses of the human face. Also, a memory based Particle Filter (PF) was used to improve the manner of prior distribution in PF and reduce the number of particles. However, a fast search method was proposed from the trained memory of PF. In fact, instead of calculating the similarity distance between each particle and the total templates in the memory, a small number of templates were utilized. The present method was tested on two available video databases to evaluate performance of proposed method. Promising results displayed better performance than the current state-of-the-art approaches in head pose tracking with our extension of the 3D Constrained Local Model (CLM-Z).