Accuracy analysis of marker-based 3D visual localization

Abstract

3D localization from images is an useful capability for robots and cameras. One successful approach is to rely on visual SLAM techniques. Another approach, maybe more robust, is to use visual markers in the environment. In this paper a study about the accuracy of marker based visual 3D localization is presented, using AprilTags markers and the solvepnp algorithm in OpenCV library. The impact of distance to markers, number of markers, their position in the image on accuracy of the 3D estimated pose is experimentally measured and analyzed.