Low cost pose estimation hardware for augmented reality processor

Abstract

After the open source tool of marker based augmented reality, ARToolKit, has been released by Kato, many researchers have tried to find creative applications using AR. That kind of works is still ongoing. Although an almost complete tool is available, one restriction still exists; AR hardly can be implemented in small platform like cell phone or UMPC or small embedded system. AR is composed with 3 parts Vision, Pose Estimation and Graphics. The critical reason which blocks AR to be implemented in low cost is Pose Estimation part. To solve that problem, low-cost dedicated hardware for pose estimation is proposed. If this is combined with mobile graphics processor which can handle vision and graphics processes, Augmented Reality could be implemented in low cost. Therefore, the development of AR applications would be activated more. This work is based on ARToolKit. For low cost implementation, some part of algorithm is substituted with another algorithm and another some part, modification, is substituted with proposed method. By the proposed modification algorithm, final result of pose estimation has 35% less error and the processing time of modification is reduced to averaged 24%. This proposed pose estimation hardware is implemented and synthesized in 0.18um technology. The size is about 140,000 gate counts at 100MHz clock frequency.