A vision-based AR registration method utilizing edges and vertices of 3D model

摘要

The most important issue in AR (Augmented Reality) is accurate geometric registration between real world and virtual objects. We propose a hybrid registration method utilizing the edges and the vertices of a 3D model of the target object. We estimate camera position and orientation by detecting the vertices and true edges every frames. The influence of the misleading edges is reduced by considering multiple edge candidates and selecting the best one from them. Furthermore the magnetic sensor or a set of the artificial vision markers as a tool to obtain approximate camera position and orientation are also used when the tracking object goes out of the user's view point or the camera movement is too fast to track natural features. By considering multiple edge candidates and utilizing a magnetic sensor or vision markers, the accuracy of registration and the robustness of rapid camera movement can be improved.