Abstract: ‍ ‍Group target tracking holds significant research value as a crucial step in countering drone swarm threats. Compared with traditional multi-target tracking algorithms， Random Finite Set （RFS） filtering demonstrates substantial advantages in handling multi-target data association and state estimation. However， in low， slow， and small group target scenarios， existing RFS filtering methods generally fail to account for the impact of group characteristics on tracking， nor do they address the challenges of group target initialization and track information extraction. To address these issues， this paper proposes a tracking method for low， slow， and small group targets， combining group modeling， a labeled Cardinality Balanced Multi-target Multi-Bernoulli （CBMeMBer） filter， and adaptive target birth intensity techniques. First， the group target structure was modeled using the virtual leader-follower model and an undirected graph adjacency matrix. Subsequently， a three-level priority label assignment strategy was introduced to improve the traditional labeled CBMeMBer filter， to solve the track ambiguity caused by label conflicts and enhance both tracking accuracy and algorithm efficiency. Additionally， an adaptive target birth intensity algorithm based on the group target scenario and a two-point initialization method was designed to achieve adaptive initialization of new group targets in the RFS framework. Finally， simulation experiments and experiments based on real measured data from the holographic staring radar demonstrated that the proposed method excels in target state estimation and track quality. The tracking performance was superior to those of comparison algorithms， including the traditional labeled CBMeMBer filter， effectively avoiding track crossing and ambiguity， thereby fully showcasing its potential and practical application value in the fine tracking of low， slow， and small group targets.