Making decision for the next step in dense crowd simulation using support vector machines

This paper presents a method to optimize the travel time of an intelligent agent finding its way through a simulated dense crowd by using Support Vector Machines (SVMs). While prior studies have used Social Force Model (SFM) for crowd simulation, a limitation of the model is that it does not allow the agent to make decision in getting to a specific point in a dense crowd in the shortest time possible. The next step refers to the temporary position before a final destination is reached and identifying a suitable next step in a dense crowd is challenging as there are many other moving agents that might block the path. Hence, to avoid formulating the decision of the agent's next step to proceed to a target point in an explicit manner, we therefore resort to SVMs whereby the position and velocity of the nearby agents are used as the feature vectors and the next step position as the class label. The training data is trained a priori with feature vectors and class labels supplied by human subjects. In order to produce a higher accuracy of training model, the training data is duplicated by permutation process. During the simulation, we used SFM and multiple SVM training models to control the motion of the intelligent agent. The results of the simulation indicated that the combination of SFM+SVMs enabled the intelligent agent to reach the final destination faster than using SFM on its own.