The Research Of Tracking Registration In Augmented Reality Based On Vision

Augmented Reality(AR) refers to adding virtual computer-generated information to the real scene, which realizes the enhancement to the real world and makes the user experience a true new environment that has a true sensory effect. AR is a new technology developed on the basic of Virtual Reality(VR), which has a very wide application prospects in the fields of military, education, architecture, medical research, engineering, industrial maintenance, tourism culture and entertainment, et al. AR has been a hot research problem in the field of computer science. The 3D registration technique is the foundation and key of the AR research, which determines the output effect of AR system mixing the real information and the virtual object together and whether the AR system is success or not.This paper mainly researches on the 3D registration of AR and proposes a novel registration method based on natural features for augmented reality.In order to meet the robustness and the real-time performance for AR system, this paper studies the technique of feature extraction and matching, and proposes to use FREAK operator to extract and match the natural feature points of images. The robustness of FREAK is better, and its feature descriptor is a binary bit string. It can be quickly calculated and generated, which can greatly reduce the time of image processing stage and improve the system speed. The matching technology is used to match the feature points and the outlier remove is used to getting the best matches. Moreover, in order to more accurately estimate the pose information of the camera and improve the accuracy of registration, this paper refines the homography that has been estimated by the best matches, the core idea of which is to conduct a perspective transformation to the current scene images. The camera intrinsic parameters can be obtained by camera calibration technique. Then the camera pose can be calculated by the camera intrinsic parameters and the homography. Finally, the virtual object can be accurately added to the real scene according to the camera pose and output the scene mixing the real information and the virtual object together.In this paper, the experiment platform is Visual Studio 2010. The experiment images are collected by the ordinary PC camera and the image resolution is 640 x480. Experimental results show that in image processing stage, the frame rate is about 24.86 FPS, which greatly reduce the processing time of registration. This method has a good accuracy and it is robustness under the circumstances of the target object partially occluded, the changes of the rotation, the scale, the ambient light and the distance.