Abstract: In conventional multiple hypothesis tracking (MHT) algorithm, a target is assumed to generate one measurement independently. In practical scenario, however, closely spaced multi-target may be identified as one unresolved measurement due to limited resolution. This phenomenon complicates the data association problem and badly degrades the tracking performances. In order to solve this problem, an improved greedy randomized adaptive search procedure MHT (GRASP-MHT) algorithm is proposed. To adapt to scenarios may contain unresolved measurements, the new algorithm derived the score of track hypothesis associated with unresolved measurements and modeled the complex data association problem as a maximum weight independent set problem (MWISP). Simulation results demonstrate that the improved GRASP-MHT can solve the data association problem with unresolved measurements and retains most of the advantages of GRASP-MHT.