Using Traffic Signs as Landmarks in Object-oriented EKF-SLAM

Simultaneous localization and mapping (SLAM) is a well-known problem in robotic applications. Many solutions consist in using handcrafted features as landmarks such as SIFT, SURF, ORB… These features are usually edges or corners which are invariant to some specific transformations. However, because of their low-level nature, they are not only non-informative but also not robust under various conditions (lightning, weather, point-of-view) and their track is often lost from one frame to another. To tackle this issue, the main idea of this work is to detect and localize higher-level landmarks such as static semantic objects instead. This paper focuses on the integration of circular traffic signs landmarks into a complete Extended Kalman Filter SLAM framework. The main part of this work consists in the quadric parametrization of the observation function and its inclusion in the Bayesian approach.