| Head pose estimation plays an important role in the face recognition, and it is also important in drive fatigue and other areas.Head pose estimation is a method which uses the real time images obtaining from the camera or the images sequences in a video to find faces and estimate the head pose. Main Method to detect faces is to use Haar features, or use the infra camera so that the light influence can be reduced. Main method to track feature points is to use the gray information. Main method to estimate head pose is to use the feature points on the face or use the dual camera to get the3D head model and then estimate the head pose.Head pose estimation includes face detection, feature points detection and tracking. This thesis focuses on the face tracking method by using the ASM technology. The MUCT face database is adapted. By annotation and training, we have a shape model and a patch model, and finally generate an ASM model. According to the requirement of the ASM model’s initialization, this thesis trained and get the relationship between the point vectors and the border of the face and initialize the ASM model by detecting the border of faces. The method shows good performance in the frontal faces situation. According to the drift in the tracking, this thesis proposes an algorithm which combines the Adaboost detection and ASM model. The experiment result shows the algorithm improves the robustness when the face disappeared or was obstructed that can reinitialize quickly. By the feature points obtained by the ASM model, the head pose can be estimated in three degrees of freedom and this thesis improve the performance of the angle estimation by rotate the image until the Theta’s angle is0. This thesis propose a method which can use the OpenCV in LabVIEW and implement the head pose estimation on the LabVIEW platform. The head pose is used to control the cloud terrace PTS-302which has a camera on it so that we can control the video angles by using the head pose. |
PDF Full Text Request