The traditional centroid group tracking (CGT) algorithm estimates group collective movement by tracking the centroid of group. However, the space distribution of group targets and the centroid of group will be influenced by clutter, result in increasing tracking errors. The improving method proposed in this paper makes good use of characteristics information of electromagnetic radiation acquired by sensors and fuses the information with the motion status, time domain, and frequency domain characteristics information of target. By calculating the degree of association between the qualified measurement and the prediction measurement, the clutter is eliminated and the status of centroid is estimated. Numerical simulations show that the new algorithm outperforms the traditional algorithm on the root mean square errors and average effective measurement quantity, resulting in the improvement in tracking performance.