REGISTRATION ERROR ANALYSIS FOR AUGMENTED REALITY

Augmented reality (AR) systems typically use see-through head-mounted displays (STHMDs) to superimpose images of computer-generated objects onto the user's view of the real environment in order to augment it wi th additional information, The main failing of current AR systems is t hat the virtual objects displayed in the STHMD appear in the wrong pos ition relative to the real environment. This registration error has ma ny causes: system delay, tracker error, calibration error optical dist ortion, and misalignment of the model, to name only a few, Although so me work has been done in the area of system calibration and error corr ection, very little work has been done on characterizing the nature an d sensitivity of the errors that cause misregistration in AR systems, This paper presents the main results of an end-to-end error analysis o f an optical STHMD-based tool for surgery planning, The analysis was d one with a mathematical model of the system and the main results were checked by taking measurements on a real system under controlled circu mstances. The model makes it possible to analyze the sensitivity of th e system-registration error to errors in each part of the system, The major results of the analysis are: (I) Even for moderate head velociti es, system delay causes more registration error than all other sources combined; (2) eye tracking is probably not necessary; (3) tracker err or is a significant problem both in head tracking and in system calibr ation; (4) the World (or reference) coordinate system adds error and s hould be omitted when possible; (5) computational correction of optica l distortion may introduce more delay-induced registration error than the distortion error it corrects, and (6) there are many small error s ources that will make submillimeter registration almost impossible in an optical STHMD system without feedback, Although this model was deve loped for optical STHMDs for surgical planning, many of the results ap ply to other HMDs as well.