Video analysis architecture for enhancing pedestrian and driver safety in public environments

One key goal of current Computer Vision research activities is to provide robust systems for improving Transport safety through the use of Information Technology. Recent advances allow public environments (such as train stations or, simply, the street) under video surveillance to be modelled by means of the detection, tracking, and identification of the different elements in it (passengers, road, vehicles, pedestrians or other objects). At a higher level, this information can be processed to detect automatically dynamic events in video sequences, which will help the system to understand what is happening in the scene being analyzed and launch an alarm in case of any anomalous behaviour may happen. In our work, we present a robust video analysis architecture, at two levels, able to process low level tasks (key point detection, movement detection, and tracking) and high level behaviour inference (object modelling, actions identification and activities classification). Basic image processing task is then complemented with the required intelligence either to accept the definition of allowed or forbidden events, or to dynamically learn about usual behaviours and anomalous ones. Experimental results demonstrate the effectiveness of the proposed system, allowing the mobility of users become more and more secure.