Automatic security system for recognizing unexpected motions through video surveillance

• Main Authors: Demarinis, Flavia, Accettura, Antonella, Garzia, Fabio, Cusani, Roberto
• Format: Conference Proceeding
• Language: English
• Subjects: Arrays; Artificial neural networks; automatic security system; intelligence; Neurons; Security; smart video surveillance; Training; unexpected motions; Video surveillance
• ISSN: 1071-6572; 2153-0742
• DOI: 10.1109/CCST.2014.6987025

This paper deals with a study concerning the so called "smart video surveillance" system, starting from the consideration of unexpected motions. It is known that security staff, whose aim is to watch monitors and approach in case something bad and unlawful happens, in any kind of location, can keep the attention up for no more than twenty minutes, then the concentration falls down severely. Since this decrease of efficacy, it may be helpful a support system in watching and analysing the real-time or recorded scenes. According to the state of the art, unlike several other examples of smart video surveillance techniques (3,4,5,6), this one described in this article does not focus on the images, it does rather on the velocity parameter. Given a certain scenario, certain behaviours, thereby velocities, are expected, and it is supposed they might happen. The anomalies recognition is done using artificial neural networks, which are built and trained in order to compare an array (target) of expected velocities in normal conditions of life, to several arrays (input) of other velocities extracted from unexpected situations properly chosen. Once decided the tool for tracking the motions in the videos, then obtained the arrays of x and y velocities components, it is the time to build the artificial neural networks through iterations, which have been done changing the number of hidden neurons. The results are interesting enough and go right to the purpose, which consists in the choice of the best neural network.