Crowd analysis by using optical flow and density based clustering

Abstract

In this paper, we present a system to detect and track crowds in an image sequence captured by a camera. In the first step, we compute optical flows by means of pyramidal Lucas-Kanade feature tracking. Afterwards, a density based clustering is used to group similar vectors. In the last step, a crowd tracker is applied to each frame, allowing us to detect and track the crowds. The output of the system is given as a graphic overlay, i.e. arrows and circles with different colors are added to the original images to visualize crowds and their movements. Evaluation results show that the system is capable of detecting certain events in the crowds, such as merging, splitting and collision.