The recovery of the 3-D movement of the face is an important operation for many applications like human machine interaction, video surveillance, MPEG-4 compression etc. This paper presents a method to obtain from a video input, a normalized face in a frontal pose, by recovering the fullmotion of the head using 3D head model. From some characteristic face points given on the first frame, an approximated 3D model of the face is reconstructed. Using this model, the full motion of the head is computed automatically. Evidently, in order to compensate errors due to the rough 3D model, a combination of several techniques has been used to reach a strong robustness. The algorithm has been tested on synthetic videos and it has been compared with a standard multi-camera system for the 3D tracking (Elite 2002 System). The results in both cases are good. The proposed approach is part of a facial expression analysis system. Our aim is to detect the facial expression in situations characterized by a moderate head motion. For this reason head motion recovering is fundamental. Once recovered the pose, we are able to obtain frontal normalized facial image that makes expression analysis easier.
Accurate 3D model based face tracking for facial expression recognition / M. Anisetti, V. Bellandi, F. Beverina - In: Proceedings of the fifth IASTED international conference on visualization, imaging, and image processing : september 7-9 2005, Benidorm, Spain / [a cura di] J.J. Villanueva. - Anaheim : Acta press, 2005. - ISBN 0889865280. - pp. 93-98 (( Intervento presentato al 5. convegno IASTED International Conference on Visualization, Imaging, and Image Processing tenutosi a Benidorm, Spagna nel 2005.
|Titolo:||Accurate 3D model based face tracking for facial expression recognition|
ANISETTI, MARCO (Primo)
BELLANDI, VALERIO (Secondo)
|Parole Chiave:||Candide-3; Expression analysis; Face tracking|
|Data di pubblicazione:||2005|
|Enti collegati al convegno:||IASTED|
|Tipologia:||Book Part (author)|
|Appare nelle tipologie:||03 - Contributo in volume|