The B-spline Gaussian mixture probability hypothesis density (BS-GM-PHD) filter can track an unknown number of extended targets and estimate their shape. However, the target tracks might be inaccurately associated when targets are closely spaced. This paper presents a novel track association approach for the BS-GM-PHD filter using the estimated shape information. First, a shape table (ST) will be established to keep the shape information of each estimated target. Then, targets will be identified by using the ST. the simulation results show that the proposed approach can accurately associate tracks, even though the targets are closely spaced.