A COMBINED TRACKING-RENDERING METHOD FOR 3D OBJECT RECOGNITION IN AUGMENTED REALITY

摘要

Most of synthetic Augmented Reality Environments are using 2D patterns laying on 3D objects for both real object recognition and tracking. But a such approach seems to be not just comfortable in every day work due to great difficulties in pattern positioning onto 3D objects. An original conceived approach for spatial positioning and recognition of real objects (AR-Ctr -Augmented Reality Combined tracking rendering) is being presented in this paper. Virtual Reality devices, like magnetic 3D tracking system provided by Polhemus (Fastrak), pinching gloves (featuring electrical contacts to activate commands and various events in virtual environment) and 3D stereo visualization may help user to overcome pattern positioning on real objects. AR-Ctr user, wearing a Sony Glasstron see-through head-mounted display, equipped with a Logitech QuickCam, is additionally tracked on both hands and head by 3 Fastrak receivers, returning into the virtual environment the relative positions of the right and left index fingers. Furthermore virtual camera and user head are collimated thanks to the third Fastrak receiver data, avoiding the time-consuming routine for virtual camera position calculation. The AR-Ctr recognition features are based on a parallel rendering task, providing a fast and camera collimated synthetic picture of every 3D model stored in general database (of all recognizable models). Bounding box captured from tracking and pinching activity gives the necessary data to match real and virtual environment. That procedure has been proved an remarkable step ahead towards complex 3D object recognition, arbitrary positioned in the space.