AI detects firearms in live CCTV feeds and alerts police
Joe Levi works hard to prevent another mass school shooting tragedy.
He says his technology, developed to detect a weapon with existing CCTV cameras, could make a world of difference in future life-or-death situations.
In 2018, 17 people died when a 19-year-old student opened fire at Stoneman Douglas High School in Parkland, Florida, USA.
Fifteen died in the 1999 Columbine High School massacre near Denver, Colorado, when a 17-year-old and an 18-year-old classmate were shot.
And 22 people died in May this year when an 18-year-old rampaged through Robb Elementary School in Uvalde, Texas, in one of the worst school shootings in US history.
Levi has developed artificial intelligence that uses existing camera networks to immediately alert police when it detects a firearm, sending them an image, moving footage and an exact location.
“The first tragedy is that 19 children and two teachers were killed in Uvalde, Texas,” he tells NoCamels. “The second tragedy is that the whole event could have been prevented.”
He says his technology would have made a “huge difference” if it had been installed on cameras around the school.
“It could have recognized the shooter with an assault rifle as soon as the CCTV cameras picked it up,” he says.
“Unfortunately, since our system was not in place, the shooter was able to enter the site and start shooting.
“Four and a half minutes later, when police got to the scene, they had assumed it was just a simple shooting and they didn’t take into account the fact that the shooter was armed with an AR-15 assault rifle.
“Because of the first responders’ lack of situational awareness, they were simply outmatched, resulting in the deaths of all those children and teachers.
“Our system leverages the existing surveillance camera infrastructure,” says Levi, who founded 1702ai, which is based in Kfar Saba, central Israel, and has offices in Zurich, Switzerland.
“And if it sees a handgun, it simply shares a JPEG image, a video clip, and the GPS coordinates of the alert directly with law enforcement.”
“The issue we are addressing is mass school shootings in the United States. Take the Uvalde school shooting, which is one of the worst in recent memory, on the scale of the Columbine or Parkland school shootings.
“You have a guy there who shoots a machine gun from outside the school for two minutes. It is captured by cameras. He enters the school, captured by cameras. He goes into a classroom, shoots for two minutes.
“The police come and have no idea what’s going on. They hear gunshots, they walk down the hall, then realize they’re just outmatched.
“It takes 80 minutes and 374 police officers to neutralize a guy. Why? They don’t have what we call situational awareness, they have no idea what to expect.”
Levi says the technology he’s developed isn’t just super accurate at detecting weapons. It also immediately alerts local people who can take action. And unlike other weapon detection systems on the market, it produces only a tiny number of false alarms, according to him.
“Even with the latest machine learning and AI algorithms, it’s difficult to detect a weapon with CCTV cameras because the camera angle and the environment are never the same and the lighting is always different,” he says.
“Some countries can be very cloudy, snowy or rainy. We work exclusively in the infrared range, i.e. only with night vision.
“I’m from Hollywood, I’ve worked on hundreds of commercials, films, reality TV, I’ve even worked as an imaging specialist for Apple. So I know what works and what doesn’t.
“In Padova, Italy, we processed 9 million images and received five false alarms. Our competitors have admitted that they have thousands of false positives.
“We have something unique in the way we learned to recognize a weapon. Nobody wants a system where a cell phone sets off an alarm or a banana sets off an alarm.”
The 1702ai technology is operational in Oslo, Norway, in Padova and at a very large transport hub in Europe that cannot be identified.
“We’re starting to work with two security companies in America, one of which handles security for about 50 universities. At the moment we are only concentrating on schools, because our first customers were governments all over Europe,” says Levi.
One of the keys to maintaining high accuracy is constant updates to the AI. “First of all, artificial intelligence has to be very accurate, but you have to keep it accurate.
“By sharing data from twin cities, we can prevent what’s known as AI drifting, which is a very common problem, but it needs to be addressed.”
When it comes to AI drifting, the technique works almost perfectly in the lab, but not so well in the real world. Levi says it’s important to prevent accuracy drift.
“You see AI in the lab in a closed environment and it works well, but as soon as you use it in a different environment, the predictions won’t be the same,” he says.
“The level of trust that was maybe 96 percent will drop to 80 percent. And then you will have many false alarms. We need to be very, very accurate with the weapon detection tool, so we’re constantly updating it, a bit like updating an app on your phone.”
Incidentally, the company’s name – 1702ai – was chosen because the first data set it looked at contained 1,702 images of real knife attacks.