Indoor Parking Localization Based on Dual Weighted Particle Filter

For successful autonomous valet parking, accurate knowledge of vehicle location in the global map is crucial. Using sensors and maps can be an alternative method for the indoor parking localization to compensate for lack of GPS coverage. However, maps only have static elements such as walls and pillars and semi-static elements such as parked vehicles are not included. Thus, sensor data rarely match with the map data due to the semi-static elements. We developed a robust localization algorithm using a laser scanner, a static map and a feature extraction algorithm. Overall map features consisted of the center positions of parking slots and pillars. Parking slot measurements were extracted from parked vehicles. Position estimation errors occur when matching the vehicle center positions to the parking slot center positions. This error can be reduced by detecting pillars and giving more weight to the observation of these static elements. The contribution of this approach is that we can use not only static objects but also semi-static objects and that position estimation error is reduced. The algorithm was evaluated in the scaled down indoor parking model. The average position errors of this algorithm are compared with errors of odometry data and SIS particle filter.