Abstract
Camera positions and attitudes are necessary to synthesize and compose images of virtual objects in AR applications. Although markerless AR techniques using local sensors such as a GPS and a geomagnetic sensor are useful without specific preparations in the AR locations, accuracies in directional sensing limit their applications. In this study, we utilize both global sensors, such as GPS and Kinect, and local sensors, such as acceleration sensors and gyroscopes. By matching the two motion paths of a given device obtained by the both global and local sensors, we can estimate the camera attitude. Simulation experiments demonstrated positive results.
