Consistency of the monocular EKF-SLAM algorithm for three different landmark parametrizations

We benchmark in this article three different landmark parametrizations in monocular 6DOF EKF-SLAM. These parametrizations are homogeneous points (HP), inverse-distance points (IDP, better known as inverse-depth), and the new anchored homogeneous points (AHP). The discourse used for describing them is chosen to highlight their differences and similarities, showing that they are just incremental variations of ones with respect to the others. We show for the first time a complete comparison of HP against IDP, two methods that are getting popular, and introduce also for the first time AHP, whose description falls precisely between the other two. The benchmarking is done by running all algorithms on the same data and by using the well-established NEES consistency analysis. Our conclusion is that the new AHP parametrization is the most interesting one for monocular EKF-SLAM (followed by IDP and then HP) because it greatly postpones the apparition of EKF inconsistency.