Exploring video data by simply watching does not scale for large databases.This problem arises in a wide variety of domains, such as scientificvideo analysis, digital life streams, and video surveillance.

Motivated by a mini challenge of the contest of the IEEE Symposium on Visual Analytics Science and Technology 2009 (Detecting the encounter of persons in a provided video stream utilizing the techniques of visual analytics), we propose an approach for fast identification of relevant objects based on the properties of their trajectories.