Science and Technology

One step ahead of violent intruders

New technology could help workplaces and schools identify violent intruders before they enter the door.
Prototype NVidia TX2 module

A small prototype created using an NVidia TX2 module. Photos: TJ Lievonen/News@TheU

Imagine if there was a way for cameras to identify violent intruders and alert law enforcement instantly, before a potential shooter even entered a school or a place of business.

This may soon be possible if a team of researchers from the University of Miami’s College of Engineering, including graduate student Saad Sadiq, junior Marina Zmieva, and Professor Mei-Ling Shyu, perfect a novel computer program that uses security cameras and facial recognition technology to identify criminals and potentially violent intruders before they enter a facility.

The algorithm they have designed—which uses artificial intelligence to recognize faces of images entered into the program and determines if a threat exists—has an accuracy rate of 97 percent. AI, which is used in Ok Google, Amazon’s Alexa, and Tesla vehicles to avoid accidents, learns tasks and is taught to react like a human. 

“Right now, the police only get a call once the shooting starts,” said Sadiq, who is a father of two. “With this device, they would still have time before the shooting starts.”

Researchers from the College of Engineering
Graduate student Saad Sadiq, College of Engineering Professor Mei-Ling Shyu, undergraduate assistant Marina Zmieva.

Some of the advantages of their idea include its portability—the computer chips they use for the program are very small and can be embedded in a Closed Circuit Television (CCTV) camera, which are used in most schools, offices, and hospitals already, Sadiq said. They also work in real-time, and therefore could reach law enforcement agencies swiftly. 

Yet, what makes their design more effective than earlier facial recognition technology is the program cannot be fooled by intruders wearing disguises. This is different from previous programs, which can be undermined easily, Sadiq added. 

“Our goal was that every school should have these cameras in the vicinity, so if a shooter was near the school or walked near the school, the cameras should capture him or her and send the information to law enforcement to stop the shooter from entering,” he said.

Due to the importance of school security today, Sadiq said the team decided not to commercialize the program, but to put all of the information needed to recreate it in their paper, so that school systems or businesses could develop their own devices using the program in the future. 

“The video quality does not need to be high, so this will be very helpful and feasible for schools,” said Shyu, associate chair of the Department of Electrical and Computer Engineering. 

To run their research, Sadiq’s team created a small prototype using an NVidia TX2 module. He said the small computer chip costs about $450, and can be installed on an existing CCTV camera, or a new one.  

The only challenge to the program? Law enforcement agencies would have to provide photos of any potential intruders or criminals, so that the program would be able to identify the suspects and report them. 

“If we got this solved, the program could go live instantly,” Sadiq added.

Sadiq and Zmieva initially came up with the idea to improve on existing facial recognition technology last winter, just two days after the shooting at Marjory Stoneman Douglas High School in Parkland, Florida. They had already decided to work on strengthening facial recognition programs after they read an article about how a person had fooled the iPhone X’s facial recognition system by putting a sticker on their face. But after learning about Parkland, the two realized that this new program could serve a loftier goal.

“It sends chills down my spine knowing my kids could be in a school shooting,” Sadiq said. “What good is technology if we cannot save our children?”