Detecção de movimentos suspeitos em seqüências de vídeo

Abstract

It’s proposed in this work new techniques to automate processes in electronic surveillance systems. The input of the algorithm described is the trajectories of people captured from real filmed sequences, which are used to define pedestrian behavior patterns. The proposed model is based in two criteria to define behaviors as usual or unusual: spatial occupancy and relations among people. The spatial occupancy criterion includes a certain training period when the scene is evaluated in order to generate a database which accounts for the spatial occupancy in each scene region. Through this database, new trajectories are classified as usual or unusual. Usual trajectories are those coherent with the training database, corresponding to the most occupied areas; whereas unusual trajectories are those that occur in low spatial occupancy regions. The interpersonal relation analysis criterion employs Voronoi Diagrams in order to evaluate a set of parameters (such as distance between neighbors and other psychosocial char