Automatic reconstruction of large-scale outdoor objects like house facades is an important component of mixed-reality systems that model and visualise real world at different level of detail. The authors are involved in a project that utilises a car-mounted LIDAR to acquire a sequence 3D point clouds representing facades in a street. No GPS or IMU is used. Hundreds of point clouds need to be automatically aligned to obtain a realistic surface model of facades. In this paper, we present and compare two solutions to this complex registration problem. Our methods are based on two different, widely used techniques for registering two partially overlapping point clouds in presence of outliers. The proposed algorithms are capable of automatically detecting occasional misalignments. We analyse the operation of the algorithms paying special attention to the robustness, speed and optimal parameter setting.