A method for accurately controlling the flow of walking crowds by introducing a psychological model is proposed. In this model, we consider the effects of human memory and personal space fof sensing the density of a crowd and predicting collisions. The speed of each walker is determined from the crowd density, and collision-avoiding behaviors are controlled by assigning rules to each condition. Various walking motions are automatically generated through playback and transition of motion capture data. Our simulation controls the flow of crowds to imitate actual crowd phenomenon. This property increases the prediction accuracy of group behaviors in very crowded spaces.