3D maps including urban structures and terrain information facilitate urban searches for robots operating in rescue scenarios. However it suffers from large memory requirements and computation time to deal with point clouds collected by range sensors. It is problematic for operating rescue robots in real time. To cope with the difficulties we propose a novel 3D mapping algorithm based on voxels. The proposed geometric-featured voxel is designed to store geometric features compactly. Geometric-featured voxel maps provide compact and plentiful information about urban environments for understanding the situation of robots' surroundings. To do so we improve geometric features to extract the principal characteristics of urban structures by using local point cloud statistics. For a human operator, we also introduce a visualization method to represent the proposed voxel maps by coloring voxel's faces and edges. In the experimental result, we analyze the proposed feature performance, memory consumption and computation time in order to evaluate an ability of the proposed map.