Robust 2d augmented reality based on homography refinement and temporal coherence

Abstract

To augment real planes with virtual objects, the camera pose relative to the planes must be estimated, which can be done by plane-based camera tracking. In plane-based camera tracking, robustness and real-time capability are two critical factors but they are conflicting. To resolve the problem, we first employ a homography refinement method. Then, we use the temporal coherence of consecutive frames to eliminate the computational overhead caused by iterative refinements. It means that the camera pose estimated in the preceding frame is used as an initial camera pose of the current frame. Use of temporal coherence allows us to use more efficient feature tracking methods and thus we can achieve additional reduction in processing time. Experimental results demonstrate that 50% of the processing time can be saved while maintaining the robustness.