High accuracy localization is a basic requirement for autonomous vehicles navigation. However, in urban environments, Global Navigation Satellite Systems (GNSS) suffer from Non-Line of Sight (NLoS) signals, multipath and sometimes a limited number of visible satellites, degrading the localization accuracy. Maps with georeferenced features are a means to address this issue. In this paper, an open access map with cadastral footprints of the buildings is used for localization. Buildings are stable over time and provide visible features in cities. Using 2D footprints of the buildings provides little detailed information, but when they are matched with long range omnidirectional LiDARs, a good quality estimated pose can be achieved. We present a method that uses the Normal Distributions Transform (NDT) to match several layers of a LiDAR scan with the map. A fast filtering method based on local linear regression is proposed to extract aligned points in the LiDAR scans which filters out the largest part of the outliers before applying the NDT optimization. The performance of the approach is evaluated on real data recorded with an experimental vehicle equipped with a ground truth. The results show that this approach is able to provide high accuracy consistent with autonomous navigation tasks.