Simulations of bi-directional pedestrian flow with different direction splits and system sizes are presented based on cellular automata (CA) in this paper. It is well known that pedestrian movement is more flexible and adaptive to the dynamic conditions than vehicular flow, so four dynamic parameters are formulated to simplify tactically the decision-making process of pedestrians, which can reflect the pedestrian's judgment on the surrounding conditions and decide his her choice of action. The relationships of velocity-density and volume-density at different direction splits and on the different size system are studied and analyzed. It is found that there are phase transitions and a critical density point in the model and the relationships of velocity-density and volume-density are different in different phases. And the ratio of direction splits and system size affect the shapes of velocity-density curve, volume-density curve and the value of critical density point.