3-D annotation to images captured from a wearer's camera based on object recognition

摘要

In this paper, it is described a method for annotating the real scene image sequences with 3-D virtual objects based on scene recognition. In order to annotate the real scene with 3-D virtual objects, we have to know not only where a user is seeing from, but also what the user is seeing. Therefore, we are trying to combine simple image recognition method with a vision based video see-through mixed reality system. It is proposed the method for starting estimation of camera position and orientation using image recognition result. Our prototype system is a 3-D poster that is a real paper poster annotated with 3-D virtual objects using the proposed method. The system consists of a PC (SGI VWS540) that the algorithms are implemented on, an ultra small CCD camera and a ultra small display.