REVISTA DYNA NEW TECHNOLOGIES

The jury of the Security Forum 2017 has failed the fifth call of the Security Forum awards, which aim to promote and promote the research, development and innovation of the security industry in Spain, by recognizing those responsible for current projects of Security research, and those projects of a significant nature that can be a model and an international showcase of the vast potential of our industry.

The winner of the Security Forum R+D+i category has been the University of Granada: Real-time video fire detection system by Siham Tabik, Roberto Olmos and Francisco Herrera.

Between 2012 and 2014, the crime rate caused by firearms is very worrying in many parts of the world, especially in countries where their possession is legally allowed. The latest statistics from the United Nations Office on Drugs and Crime (UNODC) reveal that the rate of firearm killings per 100,000 population is 62.0 in Venezuela, 15.7 in Mexico, 3.9 in the United States and 0.6 in Spain. In addition, several psychological studies have shown that simply having access to a firearm increases the likelihood of violent behavior leading to it.

One of the ways to reduce the threat of violence generated by firearms is early detection of their presence with sufficient time for the agents or guards to act. In this context, an innovative and effective solution would be to provide surveillance and / or control cameras with an automatic weapon detection system.

The artificial intelligence system developed by the University of Granada team activates a warning when it detects the presence of a firearm in a scene of a video. The system is based on the use of deep learning algorithms (artificial learning models that mimic the connections of the nervous system). In particular, a model known as CNN (convolutional neural network) is used. The learning model used to detect guns has been trained on more than 3000 images containing guns.

After training, the intelligent system acquires the ability to distinguish guns from other objects held by a person.

Subsequently, when processing a video sequence it locates the presence of guns in the images and activates an alarm. The system processes five images per second, and activates the alarm when a sequence of positive images (gun) is detected.

The smart gun detection system has been shown to work well in many popular 1990s movie scenes, The World is Not Enough, PulpFiction, Mission Impossible (Rogue Nation) and Mr. Bean (videos are available on Following github repository, https://github.com/SihamTabik/Pistol-Detection-in-Videos).

When the system detects an object in a scene of these videos with a probability of being a gun greater than 70%, the weapon is highlighted with a red box including a percentage corresponding to the probability of success. Despite the poor quality of the videos used as an example, the detector provides fairly high accuracy and a number of false positives (pistol-marked objects when in fact they were not) very low in all of them.

The applications of this security technology are multiple. For example, police or security agents could find in a video scenes where guns are displayed without rewinding hours of recording. Similarly, a security camera system could trigger an alert for the presence of guns without the need for human intervention.

Thus, a jeweler who suffers a pistol assault in his jewelry would not have to risk his life trying to press a button that warns to the police because the system already would be in charge of doing it. Currently, the system focuses on detecting pistols as they are the most commonly used type of weapons in crimes, although we are working to extend it to white weapons, such as knives, knives.

The commissioning of the system is relatively simple since it only requires a simple surveillance camera, a computer to analyze the video, and a means to send the warning through an internet connection to a control center, which can be Police or a security company.

REFERENCE:

Roberto Olmos, Siham Tabik, Francisco Herrera Automatic Handgun Detection Alarm in Videos Using Deep Learning, Neurocomputing, 2017, In press. http://www.sciencedirect.com/science/article/pii/S0925231217308196