By Hans Roozenbeek
The usage of outdoor augmented reality in applications related to geo information science is limited because of a number of technological challenges, most notably the determination of the position and orientation of augmented reality devices such as smartphones. Progress in this field is limited because existing outdoor pose estimation solutions often depend on expensive data sets or proprietary algorithms. In this research, a pose estimation system is presented which improves camera pose estimates obtained from a smartphones position and orientation sensors by combining an open computer vision algorithm with open spatial data (road polygons from base registrations). Testing the system on a series of images with sensor metadata confirms that poses estimated by the system are more accurate than poses estimated by using the smartphones sensors alone. Specifically, the tested computer vision algorithm was capable of finding a better estimate of the devices heading (compass) compared to the devices sensors in 80% of all test cases. While this improved pose estimate enhances the visual quality of outdoor augmented reality scenes, its usability in actual outdoor augmented reality applications is still highly dependent on the pose accuracy requirements of such applications. A number of improvements and directions for feature research are presented to further advance computer vision based outdoor camera pose estimation.