| Augmented reality is a hot research area of computer graphics. Being a rising branch of virtual reality, augmented reality has been paid more and more attention on. Many applications have been applied to surgery, entertainment and military. Among the augmented reality, registration is the key technology. It determines the future of augmented reality. Many problems still exist in the registration procedure. This paper has done further studies on the registration problems in vision-based fiducial-tracking augmented reality.Initialization is the fundamental step of augmented reality systems. But due to the errors of the coordinates of the points using in the procedure, initialization will fail. Thus cause the failure of the whole system. Based on the traditional calibration algorithm, the paper presents a new method to calculate the parameters of the camera. The algorithm improves the veracity of initialization.Patterns recognize is the key procedure that determines whether the virtual stuff can be correctly added to the real scene. Traditional algorithm will not work when the patterns are partly covered. This paper presents a new registration method based on homography, which can work in the condition that the inner-pattern of fiducials are not visible. The paper also presents the methods using multi-fiducials and three corners. These methods can work when the corner of a fiducial is covered.This paper presents two registration methods using fiducials and texture. Using textures, registration can be done when fiducials are totally covered.Tracking is the key factor that guarantees the registration is continual and stable. Traditional algorithm uses a fixed searching window. Under the condition of rapid movement of the pattern, the algorithm will lose targets. This paper presents an adaptive tracking algorithm that can adjust the size of the searching window automatically due to the movement of patterns. Thus improves the efficiency of the augmented reality system.
|
PDF Full Text Request