Vision-based augmented reality (AR) is a technology that enhances and extends human vision by fusing images and virtual objects generated by computers. Due to the rapid development of the computer vision techniques and the great improvement of mobile platforms, vision-based AR has attracted much attention in many fields such as health, military, industry, education, entertainment and culture. However, there is still a gap between the computational capability of a mobile platform and that of a personal computer. And many existing methods cannot be used in mobile equipments. Here, aiming to three typical application scenarios, this thesis conducts a systematic research for improving the robustness of object tracking, the computational speed and the estimation accuracy of camera localization in mobile augmented reality systems. The main contributions of this thesis are listed as follows: (1) For the augmented reality applications without structure models, a fast and effective object tracking method combining a static template with a dynamic template is proposed. The method matches the features extracted in the predicted object region with the features of the dynamically updated template, in order to avoid possible lack of matching features caused by both varying viewpoints and partial occlusions. In the case of tracking failure, the method uses the static template to re-localize the object region, so that the drift arising from dynamic template tracking can be corrected to some extent. The experiments on a mobile platform demonstrate the real-time performance and robustness of the proposed method. Compared with existing object tracking methods, the proposed method is more applicable for mobile AR applications without structure models. (2) For the marker-based mobile AR applications, a new hybrid marker is designed and an image localization method based on it is proposed. The hybrid marker is a natural image with a rectangle frame. So, it inherits the advantages of fiducial marker which can be detected fast and natural image marker which enables smooth and continuous tracking. The image localization method based on this hybrid marker can not only reduce the marker detection time, but also improve the recovery along. Since the relevant region where features are extracted is confined, the adverse effect of irrelevant image regions is eliminated. Experiments show that compared with fiducial marker and natural image marker, the hybrid marker is easier for det...
修改评论