Simulating the motion of realistic, large, dense crowds of pedestrian is still a challenge for computer graphics communities. The previous main methods based individual research, in which each individual have alone plans for itself and hence. The large groups' simulation is difficult because exhibit behavior of enormous complexity and subtlety. In this paper, we improve dynamic continuum theory for group motions simulation in real-time virtual environment, which it based on hierarchical dynamic potential field simultaneously integrates global navigation with moving obstacles such as movie objects, pedestrians. Meanwhile, to avoid local collision, a method that based on individual's variable bounding box and its previous position elimination is proposed in this paper. Experiments results show improvement model can demonstrate smooth flow under a variety of conditions, and naturally exhibit emergent phenomena that have been observed in real crowds.