How Video Surveillance Is Evolving in K-12 Schools

Video surveillance has long been used in schools to document incidents and help administrators and law enforcement determine what happened after the fact. However, advances in video analytics and artificial intelligence are expanding what school camera systems can potentially do in real time.

During a recent ZeroNow webinar, “Video Surveillance in School Security,” Randy Welch, senior director for vertical growth and strategy at Motorola Solutions, Jill Renihan, education segment development manager at Axis Communications and a former school security director, and Simon Renouf, vice president of engineering at Omnilert, discussed how those changes are affecting K-12 video surveillance. All three panelists are members of the National Council of School Safety Directors (NCSSD).

The conversation covered camera placement, AI and video analytics, access control, school buses, facial recognition, privacy, policies, training, and AI gun detection.

How the Role of School Video Surveillance Is Changing

Renihan said there is still an important forensic purpose for school surveillance systems. If criminal activity occurs on or near a campus, schools want to be able to determine what happened and provide law enforcement with relevant information. But she said newer capabilities can potentially give schools information earlier.

“If analytics can tell us that somebody is coming through a perimeter gate or climbing over a fence, for example, we want to know that before the person reaches the building,” Renihan said. “That space between the perimeter and the buildings where our students and staff are gives us an opportunity to act.”

She said camera systems should be selected and designed around problems identified through a school's vulnerability and risk assessment. Welch made a similar point, saying schools should move away from simply deciding how many cameras they need.

“For every camera, ask: What are we going to use this for? What specific purpose do we want it to serve? Then design the system around those needs.”

Renouf said many existing school camera systems were originally deployed primarily for forensic purposes or situational awareness, but he is seeing increased interest in real-time detection and response.

“I don't think schools have a lack of cameras these days. I think they have a lack of attention,” he said. “That's where integrating these systems with other technologies and having a layered approach, with cameras as a foundational part of that, comes into play.”

Using AI to Address the Limits of Human Monitoring

The panelists also discussed the practical limitations of expecting people to continuously monitor large numbers of camera feeds. Renihan said most schools don't have the staffing to devote personnel exclusively to that task, while Welch pointed to selective-attention experiments that demonstrate how easily people can miss unexpected activity when concentrating on something else.

“Now imagine you've got one person sitting in a dark room looking at 20-plus monitors, all showing feeds from different cameras,” Welch said. “There's just no way a human can effectively monitor all of that.”

Analytics can call attention to activity that may require review. Welch said an alert could be displayed on a monitor or sent to a computer, phone or radio, reducing the need for someone to continuously watch numerous feeds.

The panelists also repeatedly emphasized keeping a person involved in decisions about how to respond. Renihan noted that automatically locking doors based on an alert could create problems if students outside need to get back into a building because of an exterior threat. Welch also warned about false alerts automatically triggering lockdowns.

“AI is great for giving a human information in real time and saying, ‘Bang, you have to make a decision,’” Welch said. “But the moment you take the human element out of it, you run the risk of false alerts triggering lockdowns.”

Renouf said some organizations have implemented “soft lockdowns” in which doors can be secured while a person reviews a short amount of video and determines what is happening.

“The technology takes some of the manual, human delay out of that process and lets human beings do what they're best at, which is understanding the context and making that executive decision about what needs to happen next.”

Where Schools Are Prioritizing Camera Coverage

Before discussing where cameras should be placed, Welch emphasized that schools should first focus on securing access to their buildings.

“Cameras are fantastic. They can do some amazing things, but if you're not restricting access to your building, cameras are only going to show you the bad things. They're going to show you the things you don't want to see,” he said.

When asked whether schools should prioritize cameras inside or outside buildings, Welch said the answer depends on the environment. Once the building is secured, however, he generally favors exterior coverage first.

“Make sure all of your doors can be seen. Make sure your parking lot can be seen,” he said. If budgets are limited, he said his preference is “outside first, then inside.” Interior priorities can include cafeterias, main hallways, atriums, media centers, and other large gathering areas.

Renihan agreed, pointing to parking lots and other exterior areas as important locations to monitor for incidents ranging from vandalism and fights to severe weather. She also highlighted the potential for analytics to identify changes in crowd behavior.

“If you start to see a crowd surge or the number of people gathering really starts to grow, maybe we want to increase the number of staff in that area or at least notify our security staff that something is happening and they should pay attention.”

Both also discussed Crime Prevention Through Environmental Design (CPTED), including how building design can improve visibility and make camera coverage more effective. Welch said wider, more open spaces can improve visibility and reduce the number of cameras needed, while a well-designed building can help direct people toward areas where cameras can be strategically positioned.

Renihan agreed, recalling outdoor stairwells at a high school where she worked that created hidden areas and were difficult to adequately cover with cameras.

“Trying to cover all of those little nooks and crannies with cameras is definitely more costly,” she said. “It's better to have wider, more open spaces without those hidden areas, and it's also more cost-effective.”

Video Surveillance on School Buses

School buses were another area the panel identified as changing. Renihan said cameras have been installed on buses for years, but historically the footage often wasn't available until after a bus returned and someone retrieved the recording.

“We eventually moved further into the digital age, but we still didn't have the capacity for real-time intelligence about what was happening on those school buses,” she said.

Renihan said newer systems could potentially provide real-time video during emergencies as well as live talk-down capabilities. She recalled two particularly challenging bus routes from her time as an elementary school principal. Substitute drivers sometimes didn't know the students well enough to address behavioral problems.

“If I could have looked into that bus and seen those students, I could have called them by name and said, ‘Stop what you're doing. We're going to talk about this tomorrow,’” Renihan said. “That's more impactful than having a stranger who's driving them down the road trying to address it.”

Renouf also discussed the possibility of running video analytics locally on computing devices installed on buses rather than continuously transmitting live video over a cellular connection. If analytics detected a firearm, fight, or other specified event, he said, an alert could be sent along with the bus's location to people responsible for initiating a response.

Facial Recognition, LPR, and Privacy

Facial recognition prompted a larger discussion about how schools should balance security capabilities with privacy and legal considerations. Welch and Renihan both urged schools to determine what their state and local laws permit before considering facial recognition.

Renouf said schools should also consider whether they actually need identifying information to accomplish a particular security objective.

“I don't think security automatically trumps privacy by any means, but by the same token, privacy doesn't require us to abandon useful security technologies,” he said. “I think the goal needs to be proportionality.”

For object detection such as identifying a weapon, Renouf said the system doesn't necessarily need to know the identity of the person holding it.

“Don't store information you don't need to store, and don't make it available to people who don't absolutely need it,” he said.

Welch pointed to license plate recognition (LPR) as another technology schools could potentially use to identify people or vehicles approaching campus. At a high school where students must register vehicles to park, for example, a school could potentially place the vehicle of a suspended or expelled student on a list and receive an alert if it returns.

Again, Welch emphasized that an alert should prompt human review rather than an automatic response.

“You may see that car or that person, and that should trigger a human to go out and speak to that person or at least make the decision about what needs to happen,” he said. “There are multiple ways we can restrict who enters our buildings without unnecessarily infringing on privacy.”

Renihan said schools should also consider who can access live and recorded video, how footage is being used, where it is stored, how long it is retained, and what users are permitted to do with it. Schools should also consider how original video is preserved following an incident or accident.

Welch said those decisions shouldn't be made for the first time when an incident occurs.

“You should have a board policy for every single one of those things,” he said, specifically pointing to policies for video retention, camera outages, access to footage and circumstances in which video may become an educational record under FERPA.

“Policies, if they're done well, can make your day-to-day operations run better,” Welch continued. “But when there's even a simple investigation, a lack of policy can open you up to significant liability.”

Technology Selection, Training, and Ongoing Support

The panelists also addressed communication gaps between schools, manufacturers, and integrators. Welch said technology companies need feedback from schools about what they actually need and how systems are being used.

“Sometimes we go, ‘I think the world needs this.’ We design something that's really cool and then realize, ‘Oh, wait a minute. Maybe that's not what people actually need,’” he said.

Renihan encouraged schools to participate in pilots and beta testing and to be candid when a product doesn't meet their needs. Manufacturers, in turn, need to seek out and respond to that feedback. Renouf similarly emphasized testing technology with end users before deployment and continuing to evaluate its performance after implementation.

Renihan said that relationship should continue into training.

“Let's say I buy a new camera system. I get trained on it one time, maybe by my integrator, and then I'm supposed to remember how all of those bells and whistles work for the rest of my life. It's really important that we don't do that," she said. “Real learning doesn't happen in a one-time sit-and-listen or sit-and-watch session. It happens over time, with people actually using the technology, seeing how it flows and functions and how it works in their daily lives.”

The Capabilities and Limitations of AI Gun Detection

The panelists also addressed the reliability of AI-based firearm detection. Renouf said detecting a person or vehicle is generally less challenging because those are relatively large objects. A handgun, by comparison, may occupy only a small portion of the camera image.

“Cameras designed for broad surveillance sometimes don't have the resolution you need to reliably detect a gun at a great distance,” he said. “So there are some camera-selection challenges there. You need to make sure you're getting the coverage you need in the places where you need it in order to reliably detect a gun.”

Renouf said false positives are another consideration because other handheld objects, including phones and scanners, can resemble firearms from certain angles. He said Omnilert uses human reviewers as an initial verification step before verified alerts are escalated to a decision-maker.

Welch again emphasized human verification.

“What AI allows us to do now is, rather than having us look through hours and hours of video, it can say, ‘Hey, I think this is it. What do you think?’ Then we can make a decision from there.”

In their closing remarks, the panelists returned to several themes raised throughout the discussion. Welch urged schools to prioritize access control. Renihan recommended starting with the problem a school is trying to solve and recognizing that the solution may sometimes be training or culture rather than technology. Renouf encouraged schools to consider how individual technologies fit into a layered security plan and to select systems that can work together.

Watch the full ZeroNow webinar →

School Security Leadership subscribers can also access the complete edited transcript of “Video Surveillance in School Security,” organized by topic with timestamps corresponding to the webinar recording.

Access the Full Webinar Transcript →