When the emergencies happen, the pedestrian walking behavior would change by the emergencies and their influence. The research on the pedestrian walking characteristics in emergencies could optimize the evacuation efficiency. Aiming at the shortcomings of the data acquisition in the existing research, the real pedestrian scene video was processed and the relevant data were extracted, and the general walking characteristics of the pedestrian without emergencies were analyzed. Aiming at the pedestrian flow in emergencies, the spread of the emergency influence and the pedestrian self-organization phenomena in emergencies were described by k-nearest neighbor algorithm and the resultant force, a novel Cellular Automata (CA) model whose cellulars were suffered from the influence of normal walking, emergencies and security marks was proposed. The simulation of bi-direction pedestrian evacuation in emergencies was carried out by the proposed model. The experimental results show that, when the separation distance of security marks in a narrow pedestrian passage mark is 0, 10, 20 cellular, the distance of the safety mark distribution has not obvious effect on pedestrian evacuation. Through the study of whether there is influence among the population, the effect of evacuation is mainly affected by the spread of emergencies by nearby pedestrians. That the impact of emergencies is too large or too small will cause congestion and is not conducive to the evacuation of the crowd. The simulation results are consistent with the scenario of pedestrian evacuation in reality.