This paper considers multiple resolvable group target estimation under clutter environment. The proposed algorithm involves two aspects: target estimation and group state (group size, shape, etc.) estimation. First, we propose dynamic models and observation function for the group targets. Second, we derive the connection relation of individual targets through the predicted target states. In the following step, we combine the graph theory with the group targets and build the adjacency matrix of the estimated state set. The connection information is used to correct the collaboration noise and estimate the target states. For group estimation, we focus on the number of subgroups, the group states and the group sizes. Finally, several examples are given to verify the proposed algorithm, respectively. (C) 2017 Elsevier Ltd. All rights reserved.