For mobile robot clusters, a ground-end Lidar localization system needs a small number of Lidar sensors. However, it has several technical problems, such as severe feature loss, difficult target segmentation, etc. Hence, a multi-target tracking method is proposed, which is based on hierarchical segmentation and clustering of multi-perspective Lidar point clouds. Firstly, a dynamic threshold segmentation technique is used to convert multi-source Lidar point clouds into point groups. A surface area heuristic-bounding volume hierarchy method is devised to distribute these point groups into point cloud space, and a spatial hierarchical tree is established using the recursive algorithm. Secondly, the loss function of distance-intersection over union (D-IoU) is improved, and a relevant point group clustering method based on the undirected graph is proposed to generate the target point cloud for each mobile robot. Thirdly, the Kalman filter is used to track the target point cloud of each robot. A depth-first searching technique based on the spatial hierarchical tree is proposed to track the target point cloud. Finally, a multi-perspective ground-end Lidar tracking system is developed. Target tracking experiments are implemented for a group of mobile robots, which evaluates the effectiveness and efficiency of the proposed method.