Vision-based registration for augmented reality-a short survey

摘要

The purpose of this paper is to explore some existing techniques in vision-based registration for Augmented Reality (AR) and present them collectively. AR is a branch of computer vision which generally overlays Virtual Objects (VOs) on actual images of real-world scenes in order to provide additional information about the scene to the user. Due to its wide range of applications in the fields of medical, robotics and automotive, geographic and remote sensing, military and aerospace, it has gained high demand. In any AR system, registration is the key to make the augmented scene appearing natural. Registration process must avoid occlusion of VOs and objects in the real world and align the VOs precisely. Optics-based and video-based are two well-known industrial AR systems. Researchers show that even with a single camera model registration for an AR is plausible but, VOs may be registered in front of real-world objects. It is because the registration process lacks depth information of the scene. However, employing stereo vision system and utilizing available natural features in a real-world scene and set of arbitrary multiple planes one can improve accuracy of VO registration. Thus, an AR system becomes robust if it is devised with algorithms to extract and track natural features in real-time.