Pattern Recognition in Spatial-Temporal Graphs



We model crowds as dynamic spatial graphs with a node representing a person, and a link representing the fact that two people have been detected to be in each other’s proximity. The result is a continuously changing graph. When observing crowds in terms of such graphs, patterns emerge such as lane formation, cyclic movements, congestions, and so forth. In this project, we study the automated detection and recognition of such patterns from dynamic spatial graphs.