Trajectory-based motion pattern analysis of crowds

Various techniques have been developed in recent years to simulate crowds, and most of them focus on collision avoidance while ignoring basic statistical spatiotemporal properties that crowd should possess. In order to improve the quality of crowd simulations, in this paper, we investigate some statistical characteristics of pedestrians in unstructured scenes using captured motion trajectories. Each trajectory is first represented as a four-dimensional vector, following which trajectories with the same entrance/exit areas are clustered to form motion patterns using the fuzzy c-means (FCM) algorithm. Since errors arise during tracking, outliers are eliminated using the local outlier factor (LOF) algorithm, and the refined velocity field can then be obtained. Finally, for each motion pattern, we find and confirm the following three spatiotemporal statistical properties of pedestrians: 1. The distribution of path length obeys the power law. 2. Pedestrians' speeds follow a Gaussian distribution. 3. Pedestrians tend to maintain a lower speed in entrance/exit areas and a higher one in the middle of a given path.