Advanced real-time indoor parking localization based on semi-static objects

Indoor parking localization systems face two major challenges: the absence of GPS coverage and the dynamic character of the parking environment. The absence of GPS coverage can be compensated by data fusion of several sensors in order to localize an object in a map of the environment. However, this map may constantly change due to the dynamic nature of a parking garage. This paper describes an improvement to indoor localizations by a new map representation approach. The algorithm is developed for the application of an indoor parking system. The car park's map is based on the known permanent elements of the environment in combination with the knowledge of possibly changing elements. The permanent elements (e.g. walls) are considered static objects, while the known changing elements (e.g. parked cars) are modeled as semi-static objects. Such a representation avoids constantly updating the map with newly detected objects. Our algorithm fulfills the real-time requirement of the localization task for a computer-controlled car through the use of a particle filter. The filter uses laser scanner measurements and odometry data to localize the car in a precalculated probability grid map, containing static and semi-static elements. The evaluation of the algorithm in a real car park scenario demonstrates the robustness and real-time capability of the proposed system with a RMS error in the absolute position of 0.33m and in the heading angle of 1.03° at a computation time of 10ms per cycle.