A3FD: Accurate 3D Face Detection

Face detection has recently received meaningful attention, especially during the past decade as one of the most prosperous applications of image analysis and understanding. Video surveillance is for example, one emerging application environment. This paper presents a method for accurate face localization through a coarse preliminary detection and a following 3D refinement (A3FD: Accurate 3D Face Detection). A3FD can be useful applied to video surveillance environment thanks to the performance and the quality of the results. In fact for many application (e.i. face identification) the precision of face features localization is a real critical issue. Our work is therefore focused on improving the accuracy of the location using a 3D morphable face model. This technique reduces the false positive classification of a face detector and increases the precision of the positioning of a general face mask. Our face detection system is robust against expression, illumination and posture changes. For comparison purposes we also present some preliminary results on largely used face database.