Dynamic weighted aggregation for normality analysis in intelligent surveillance systems

Intelligent surveillance systems should be able to carry out an exhaustive analysis from multi-sensor information according to multiple events of interest in order to classify situations as normal or abnormal. That is why the design of appropriate fusion methods is essential to combine the information from a number of monitored aspects and achieve a reliable interpretation of the environment state. Unfortunately, these systems operate under highly dynamic conditions. A static configuration of the weights that determine the importance of the monitored aspects or events of interest may lead to a high number of false alarms and the ignorance of critical situations. aper performs a thorough study of different information fusion algorithms and proposes a method for the automatic reweighting of the values that establish the importance of the analyzed events of interest. This online method is flexible enough for adjusting such weights in each monitored situation to address the dynamic nature of real environments. The experiments, which have been conducted in a real urban traffic environment, demonstrate the feasibility of the proposed method.