Fast surface reconstruction from dense point clouds data(PCD) is a popular topic in robotics. However, it is still challenging to make large-scale surface maps with sparse PCD. This paper proposes a method to build surface maps using the PCD generated by velodyne® HDL- 32E laser scanner, which is wildly used in automatic driving vehicles and wild exploring robots. Supervoxel clustering method is employed in this paper to organise the spare point clouds, and scale variable triangular meshes are generated to represent the large-scale outdoor scenes. Experimental results show that the proposed method is efficient and robust even in uneven terrains contain complex obstacles.