A System for Visual Analysis of Objects Behavior in Surveillance Videos

Closed-circuit television (CCTV) surveillance systems are employed in different scenarios to prevent a variety of threats, producing a large volume of video footage. Several surveillance tasks consist of detecting/tracking moving objects in the scene to analyze their behavior and comprehend their role in events that occur in the video. Such analysis is unfeasible if manually performed, due to the large volume of long duration videos, as well as due to intrinsic human limitations, which may compromise the perception of multiple strategic events. Most of smart surveillance approaches designed for moving objects analysis focus only on the detection/tracking process, providing a limited comprehension of objects behavior, and rely on automatic procedures with no/few user interaction, which may hamper the comprehension of the produced results. Visual analytics techniques may be useful to highlight behavior patterns, improving the comprehension of how the objects contribute to the occurrence of observed events in the video. In this work, we propose a video surveillance visual analysis system for identification/exploration of objects behavior and their relationship with events occurrence. We introduce the Appearance Bars layout to perform a temporal analysis of each object presence in the scene, highlighting the involved dynamics and spatial distribution, as well as its interaction with other objects. Coordinated with other support layouts, these bars represent multiple aspects of the objects behavior during video extent. We demonstrate the utility of our system in surveillance scenarios that shows different aspects of objects behavior, which we relate to events that occur in the videos.