Abstract
dc:descriptionIn the field of hardware development the new technical advances provide innovative human-computer- interfaces by increasing computing power. In particular the non-intrusive mimic analysis has a high potential useful for many application areas, as e.g. language recognition, special control or medical engineering. In the called areas it is desirable to extract the essential features of human face (head-pose, the location of eye-brows, the gaze, and mouth) contactless and to interpret them subsequently. For this task there are already different procedures which show a good recognition capacity, provided an idealized framework. Varying illumination conditions and complex image backgrounds are disregarded (not considered). An extensive training is often necessary as well. Various approaches require additional special hardware, as e.g. stereo cameras or infrared-radiator, to gain additional information thereby. The presented work deals with the development of a video-based recognition system for non-intrusive analysis of facial appearance, whereas only a notebook and a webcam are applied. Compared to the existent systems a higher independence will be achieved, concerning the environment and individual features of the user. This can particularly be guaranteed by the implementation of adaptive approaches combined with a biomechanical face model. Thereby the approaches applied work in the real time, so that first-time processing rate up to 40 images per second can be achieved on the standard hardware. The concept of this work is based on so-called Active Appearance Models, which combine statistical knowledge in respect of geometry and texture information. The adjustment of face-graphs correlates highly with the underlying training material. In the existing work for the generation of training date a biomechanical 3D head-model will first time be fit, which is only based on one users frontal view. This model makes possible to simulate synthetically different illumination circumstances and face expressions. The evaluation of the complete system was carried out in the context of video-based sign language recognition, because in that the non-manual features play a decisive role concerning the information transfer. Thereby promising results had been arisen, which could also be confirmed under real conditions in form of a special control system for wheelchairs.
Degree
thesis:*- Grantor dc:publisher
- Publikationsserver der RWTH Aachen University
- Year dc:date
- 2005
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Canzler, Ulrich
- Contributors dc:contributor
-
- Kraiss, Karl-Friedrich
Subjects
dc:subject × 2Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- ger