Monocular SLAM: Using trapezoids to model landmark uncertainties

Abstract

In this paper we develop a new method to address the monocular SLAM problem based on particle filters. We already offered an approach to use line pieces attached to particles to model uncertainties of landmarks in case the measurement noise is ignorable. However, when using a consumer camera on a mobile robot, the noise resulting especially from the unavoidable vibration of the camera cannot be neglected. In this paper we present a modification demonstrating how the noise can be handled. The high performance of the algorithm is demonstrated at the end of this paper through simulations and by presenting the result of the implementation of this method on a real robot.