Human tracking based detection (HAR) in recent years has attracted much attention from the research community due to its challenges as well as wide application. In this paper, we investigate the use of Poselet detector verified by Gradient Local Auto Correlation (GLAC) and Support Vector Machine (SVM) for detecting human boundary. After that, each detected candidate is associated with a suitable tracker following the Probabilistic Data Association Filter (PDAF). For those tracker that has not been assigned, a tracking approach based block Particle Filter is in charge to maintain their existence. We conduct our experiments the open dataset PESTS 2009 and our self-recorded one CNU. We compare our method with Adaptive Boosting (Adaboost) particles filter tracking in the case of CNU dataset. The average accuracy of Blocked Particle Filter is 56.41% which outperforms that of Adaboost tracking with only 25.58%.