Performance evaluation of multi-target tracking without knowing ground truth

Multi-target tracking (MTT) has been an important and challenging area of research in the past several decades. A number of algorithms for MTT have been proposed and related performance evaluation (PE) is also gaining more attention. However, these PE methods assume exact knowledge of the ground truth, which is far from the reality. In this paper we deal with the PE of MTT without knowing ground truth based on our mock data method. The main difficulty to apply this method in the PE of MTT is the existence of clutter in the observations. We propose a method to reduce the disturbance of clutter and get a smaller data set which is more suitable for PE by incorporating the prior information about the target motion. Further, the number of targets to find the maximal possible trajectories is determined by the proposed method. Then more accurate PE results related to the number given by the MTT algorithm can be obtained. A simulation example showing the efficiency of the proposed method is given.