AI detects violence
AI surveillance systems have come a long ways, and have become available everywhere from protecting your home to investigating crimes. They can also help authorities keep tabs on high-risk areas and prevent crimes from happening.
Despite the development of AI, most public space surveillance systems still rely on human operators. However, this process can be time-consuming and may lead to a lack of detection.
For a long time, scientists have been working on developing technologies. Intelligent systems would allow public authorities to detect any unusual behavior more quickly. Combining artificial intelligence analysis modules with surveillance systems allow them to help detect unusual situations.
One key priority in the field of security and surveillance is to identify potentially violent behavior in public place. This could allow the police to intervene immediately.
A team of CUP researchers have developed a new type of content analysis software. The software is able to quickly identify when people are displaying violent behaviour based on videos. The proposed method can be more precise and cover more scenarios. The method uses a 3-D CNN to detect violent acts performed simultaneously by multiple people, not just one person.
The researchers reported in their paper. “Detecting violence in places where hundreds of people are present at the same time is particularly important, for a variety of reasons, but little research has been done in this area. Based on this situation, the only efficient way and the most important solution was to use a violence detection method based on a deep 3D neural network (3-D CNN) that would instantly extract the information in real time, and then provide it to the police to deal with the situation.”
There are two types of methods for detecting violence through video surveillance. The first type of uses traditional information extraction, while the second uses deep learning methods. Deep learning models for violence detection are more convenient and effective than traditional approaches, and this new method belongs to this category.
To train the advanced system, they used 500 videos with violence in crowded places, and 500 videos without violence at all. They made sure to conduct the research on a large scale which was necessary since they would be looking for patterns of use. CNN’s current model for detecting illegal activity was inspired by a similar network developed by the Facebook AI Lab, back in 2014.
To scientifically test their method, the scientists performed a series of tests on the Nvidia Tesla K80. This method found to be highly accurate when comparing it to three existing traditional methods.
If you liked this article, then you might also enjoy reading this one.
Author: PC-GR
The World of Technology
