The safe operation and efficient decision-making of the ground unmanned platform rely on accurate localization of the platform and overall perception of the environment. For the problem of limited perception of single-view robots, a laser SLAM (simulation localization and mapping) system based on the ground-to-air-view information fusion is proposed. Firstly, the aerial point cloud map constructed by the UAV (unmanned aerial vehicle) is introduced as the prior information in the system. Then, the optimal registration from the aerial submap to the ground local map is obtained through the registration network for ground-air point cloud submaps. After that, the aerial prior information and the ground perception information are fused based on the multi-view graph optimization framework. Finally, experiments are carried out on a road about 1000 m long in a construction site. The average translation error of the proposed system is reduced by 5.87 m compared with the classic single-view laser SLAM system, while the average rotation error is reduced by 1.67°. The results show that the proposed method effectively improves the localization accuracy of the ground unmanned platform. In addition, the perceptual blind areas of ground unmanned platforms, caused by the intersection structure, the occlusion of obstacles, etc., are effectively made up by map fusion in the proposed system.