NIST tackles video analytics for public safety
Connecting state and local government leaders
The Analytics Container Environment provides a modular framework for running containerized analytics on streaming video.
The National Institute of Standards and Technology is testing a tool that could make it easier for the public-safety community to prototype cutting-edge analytics on streaming video.
The Analytics Container Environment (ACE), being developed to support NIST’s Public Safety Communications Research (PSCR) Division, provides a modular framework for running containerized analytics on streaming video. Essentially, it could be used to detect footage of erupting fires, fights or other emergencies that demand swift public-safety response.
“If there’s a crowd suddenly gathering or a fight breaking out, or a person falls down for no apparent reason, those are important things you would want somebody to notice,” said James Horan, the ACE project lead. “Everybody says they have tons of cameras and nobody to watch them, so [this is] the idea of a detector being able to flag a camera and say, ‘Hey, there’s a fight starting on Camera 27. You may want to look at this.’”
Work on ACE started in 2016, when NIST’s Information Access Division in Gaithersburg, Md., began working with PSCR in Boulder, Colo., on the Public Safety Innovation Accelerator Program (PSAIP). PSCR was looking for projects that would support public-safety priorities, one of which is improved analytics, Horan said. ACE aims to provide a way to incorporate different standalone algorithms or systems into what he called a single testbed for comparing, testing and tweaking.
The tricky part was getting the right data to test. Because privacy is of utmost importance, the videos could not show any identifiable information, such as faces. Horan and his team started with data already in use at NIST. It wasn’t public-safety data, but it provided “a way to wring out the system,” he said. “One of the things we looked at was cameras that had just traffic -- cars that were going by that you couldn’t recognize, you couldn’t see license plates.”
So far, the team has not been able to run tests with public-safety data. Houston and Baltimore were onboard with participating, but other city needs took precedence, Horan said.
Still, testing is ongoing. The team is currently focused on incorporating analytics at the edge. The environment is designed to be scalable so it can run on a small computing platform next to a camera with some storage, Horan said.
Additionally, they can use ACE to test simulated impairments to networks, such as how analytics will be affected when a camera needs to increase compression because it’s sensing a drop in frame rate.
“In a perfect world, we would set this up in a public-safety space, and they would have a monitor that would show detections and perhaps pop-up cameras to show them what it was detecting,” Horan said. “We want to demonstrate analytics and show them how it would work because we’re trying to get both sides towards the middle: explain to the public-safety guys what the importance from the scientific side is and to the scientific side, we’re trying to explain what the operational side is.”
Another project goal was to make ACE open source so that others could use it. For example, public-safety organizations that want to fuse an object motion detector and a person detector can put those analytics in as modular pieces and connect them either sequentially or in parallel, depending on the application, and return meaningful results.
Additionally, analytics come in many forms. From PSAIP, Horan’s team received Docker containers and entire systems in Linux environments, he said. By being modular, they could wrap whatever they got in a standardized way.
“I like to call it the Swiss Army knife because you can do a lot of different stuff with it. It was modular by definition,” Horan said of ACE.
Currently, cloud is not part of ACE, but it could be, Horan said -- as long as public-safety agencies can use it. For many, that’s not realistic. When 5G arrives, he said, “we may be able to build a distributed computing environment made of ACE nodes closer to the edge and have forward-deployed analytics.”
Horan said setting up an experiment with Houston is still an option and that he has been talking with the New Jersey Regional Operations and Intelligence Center, too.