Augmented reality registration method based on natural feature points

A three-dimensional registration method based on natural feature points is presented, which is applied to the augmented reality system. Firstly, several common feature extraction algorithms have been compared in this paper, and the SIFT algorithm is applied to improve the matching accuracy as a descriptor matching method. And then the data of ten key image frames is used to reconstruct the 3D structure of the scene. When the real-time image data is input, the key frame which mostly matches the current image is selected, and the image matching method based on key frame is used to obtain the camera pose. Finally, improved Lucas-Kanade method for real-time tracking is adopted, not only maintaining the accuracy of registration but reducing the system computing time. The experimental results shown that the method achieves a effect of real-time tracking and accurate registration, and can keep a better registration precision.