Environment perception in satellite signal denied situations is one of the challenging problems for multi-robot collaboration. Visual collaborative mapping is an effective solution. However, visual collaborative mapping using Unmanned Aerial Vehicle (UAV) and Unmanned Ground Vehicle (UGV) encounters difficulties such as heterogeneous map fusion and high memory demand for saving the fused map. This paper proposes a heterogeneous map fusion method for Generalized Voronoi Diagram (GVD) and octree map (OctoMap)using point cloud as the transitional map. The UAV builds a global GVD map of the reconnaissance. The UGV establishs OctoMap of the local area after moving to the mission area according to the guidence provided by the UAV. Both the global GVD and the local OctoMap are generated according to point cloud generated by ORB-SLAM3 implemented in the UAV and the UGV. The fused map is presented in the form of high-precision OctoMap in the mission area and GVD map in the non-mission area to reduce memory cost. The proposed map fusion method is demonstrated in a simulation environment.