A study of virtual visual servoing sensitivity in the context of image/GIS registration for urban environments

This paper studies the sensitivity of pose estimation to the 2D measure noise when using virtual visual servoing. Attempting to apply virtual visual servoing to image/Geographic Information System (GIS) registration, the robustness to the noise in images is an important factor to the accuracy of estimation. To analyze the impact of different levels of noise, a series of image/GIS registration tests based on synthetic input image are studied. Also, RANSAC is introduced to improve the robustness of the method. We also compare some different strategies in choosing geometrical features and in the treatment of projection error vector in virtual visual servoing, providing a guide for parametrization.