The detailed reconstruction of complex unknown environments is a challenging task in several robotic applications. Multi-robot systems combining the advantages of aerial and ground vehicles are often used for this purpose, where the different perspectives of sensors are leveraged to reduce the proportion of uncovered areas. Since the lower speed of ground robots in cluttered environments compared to aerial vehicles can significantly limit the efficiency of the exploration, in this paper we propose a cooperative mapping system in which only the necessary regions are assigned to the ground vehicle. These regions are selected based on the structural analysis of the 3D model built by an aerial robot and a global path is computed connecting them which only passes traversable parts of the environment. A ground 3D map is also created and the two models are fused when the robot reaches the last waypoint of the path. We tested our solution in multiple simulated environments to verify its functionality.