Research On Real-time Tracking Of Augmented Reality Based On Natural Feature

Augmented reality could strengthen the understanding of the real world by registering some data generated by the computer on the natural world. Combing with interdiscipline research fields including computer graphics, pattern recognition and machine learning, it is widely used in many areas, such as military, entertainment and medical areas etc.Key technologies of augmented reality could be divided into two parts:feature extraction and registration and feature tracking. Firstly, we focus on the augmented reality using natural feature instead of the common markers. And natural feature extraction and the registration method are studied in this paper. Secondly, to solve this mismatch problem in tracking, a method is proposed in this paper based on the position of the matched points and expriment and math verification is also shown. Finally, a system is designed and implemented and the validity of proposed method is shown in this system. This research work is composed with three main sections:the first section is natural feature extraction algorithm; the second one is the natural feature tracking and registration method and the last part is design and implementation of AR systems..(1) Study of the natural feature extraction methods.The feature extraction algorithm is the foundation of an AR system Compared with other algorithms in the speed and the robust, such as SURF, SIFT, FASTSURF and FAST, the SURF and FAST is used as the extraction method. The speed of system could be accelerated with the FAST algorithm and the accuracy could be improved with robust SURF decreasing the stability caused by FAST. The experiment shows that the AR system would be more fast and stable with the robust SURF and quick FAST algorithm. Furthermore, the means which K-D tree is used to store the offline data feature to improve the match speed could also increase the speed.(2) Improvement of Natural feather Track and registration algorithm. The tracking and registration method is to get the match data of the different frames. But the registration error would be accumulated with time because of the natural feature mismatches. To solve this problem, experiment to verify the match result by straddle experiment is used to decrease the mismatch counts. We verify the method by experiments and mathematic.(3) The design and implemention AR application. AR could be applied for two kinds of environment:indoor environment and outdoor scene. A reading AR system is designed in this paper for the indoor environment and we only use the computer vision method to augmented the virtual objects to the scene. The method presented in this paper is used in the system and the experiment show it validity. A militery training AR system is implemented for the large-scale environment and some sensors, such as DR, GPS are used in the system to cooperate with the computer vision method. At last, taking the above sysytems as examples we compare and analysis these two kinds of AR application.