Parametric fusion of complex landmark observations present within the road network by utilizing bundle-adjustment-based Full-SLAM

Maps representing the detailed features of the road network are becoming more and more important for autonomous driving and next generation driver assistance systems. The mapping of the road network by specially equipped vehicles through the well-known map providers leads to usually quarterly map updates, which might result in problems encountered by autonomous vehicles in the case that the road information is outdated. Furthermore, the provided maps could lack details relevant to autonomous vehicles and next generation driver assistance systems. As an alternative, road network data can be acquired by common vehicles and subsequently fused at the backend-side to detailed, up-to-date maps. The deduction of maps consisting of point-shaped landmarks by utilizing approaches to SLAM is not trivial but well studied. However, the road network is comprised of many also non-point-shaped landmarks, such as crossroads, roundabouts, sign gantries, traffic islands and crosswalks. The reduction of these complex landmarks to point-shaped ones or the independent consideration of their basic features, such as points and lines, involves a severe loss of useful information and/or accuracy and is, therefore, not appreciated. Instead, we propose a novel approach to the parametric description of complex landmarks as a directed acyclic graph and to handle complex landmark observations by an extended formulation of Bundle-Adjustment-based Full-SLAM. We focus on the challenges imposed by the fusion of complex landmarks, such as attribute interdependencies, dynamic attribute counts, and partial observations of attributes due to occlusions. Furthermore, we show the convergence and consistency of our approach to the fusion of complex landmark observations by a hybrid, real-world scenario consisting of roundabout and traffic sign observations. The approach is presented in a way that allows a straightforward adaptation to other kinds of complex landmark observations by defining an appropriate parametric description.