Using Monocular Camera To Parallel Tracking And Mapping

Augmented reality (AR) is a process overlaying the virtual model generated by the computer on real-world scene. AR technology is widely applied in many areas, such as medical, industrial, military, entertainment and so on. This paper focus on a single camera-based parallel tracking and map building augmented reality system, unlike the majority of augmented reality systems operate with prior knowledge of the user’s environment, the algorithm of this paper do not use a prior map, and has no deep understanding of the user’s environment. Create a map in an unknown scene is a simultaneous localization and mapping algorithm. Main works in this paper can be described as follow:Firstly, the research to simultaneous localization and mapping algorithm, which use two main classical algorithm to solving this problem:the extended Kalman filter algorithm and FastSLAM algorithm based on particle filter.Secondly, the research to monocular camera parallel tracking and maps, in the study of the tracking system, an improved method is proposed, we replaced the original gray matching algorithm with optical flow method, in addition, we use bidirectional optical flow tracking method, further enhance the stability of the tracking.Thirdly, a method to improve the tracking quality testing is proposed, keep track of each frame image tracking quality testing is for early detection of tracking failure phenomenon. The feature which located in the high-level of image pyramid contains more information, so we add the top two tiers of the pyramid image feature matching ratio, and we also give more reliable threshold through a large number of experiments.