Research Of Crowd Motion Model Based On Modified Social Force

In order to simulate the experimental data, which are not available or difficult toobtain in ordinary circumstances, such as large public places stampede crowdedevent, modeling the movement of groups in the real scene more accurately is needed.The occurrence of these incidents is inherently unpredictable, and unrepeatable.People can neither wait for the happen of an accident in a given scenario, norartificially cause these incidents. Therefore, large-scale casualties’ video samples ofsuch groups are extremely rare and precious, and can’t be obtained. However, thegroup movement model to simulate the mass movement in the real scene is a methodof gaining the valuable experimental data.This paper presents a mass movement simulation model which is a kind ofsocial forces model based on group and heuristic rules. This model can simulatemany macroscopic phenomena such as avoidance phenomenon, Lane formationphenomenon and crowding phenomenon. The micro-contrast experiments show thepresent social force model based on group and heuristic rules has a smaller relativedistance error than the Moussa d’s. That is to say our model is a more realisticsimulation model of the real scene.The introduction of the concept of cohesion and group, makes the groupmovement fit in better with the law of group motion which has certain socialrelations among group members, and reduces the number of computing units toimprove the efficiency of the model calculation; correct the two heuristic rules anduse the idea of local shortest path and piecewise function to describe the individual’sdesired direction; in the process of model calculations, joining the avoidance distance,effectively solves the collision problem in Moussa d’s social force model. Calibratethe model parameters using genetic algorithms, and experiment three typicalmacroscopic phenomena and compare the relative distance error with the real scenevideo. Experiments verify the validity of the model. Experimental results show thatthe improved model has smaller relative distance error and shorter running time than the original model.