Some remarks on the statistics of pose estimation

Abstract

This work argue that directional statistics are important in Robotics. That is, statistics on general manifolds. Historically the subject began with statistics on circles and spheres, hence the title. It is still a relatively new discipline, see [4] for an overview. In order to make the case a specific example is studied: Finding the rigid transformation undergone by a camera from a knowledge of the images of a number of points. This very simple, perhaps naive, example allows us to study different models for error in the observations and estimators for the rigid transformation.