Krish Kumar, CEO at Wowza.
getty
Most businesses never set out to be in video. They didn't buy cameras because they wanted to be a media business; they needed day-to-day visibility into operations. But then feeds multiplied as more visibility and compliance were required. Missing coverage had catastrophic consequences, so eyes-on became a way to protect the business.
Now, the video drives are full, there are thousands of hours of footage sitting around, and no one is quite sure what to do with it.
This is not a niche problem. There are over a billion surveillance and operational cameras deployed globally, generating exabytes of footage every day. Most of it is never reviewed by a human and never analyzed by a machine. So, what is the point?
Most of the video that matters is not entertainment, even though there has been an explosion of entertainment-oriented content. The bulk of the video is operational. At Wowza, we support more than 200,000 active video instances across 140 countries. Our customers broadcast missions from space, stream live from NICU units and run video infrastructure in interview rooms across the country—millions of hours a year. Some of the video footage is streamed to an audience. A great deal of it is captured for compliance, safety or a record that may or may not ever be reviewed.
The questions we hear are the same everywhere: How do we do more with the video we are already capturing? How do we turn it from a cost into something the operation actually uses?
The honest answer for the last few years has been that AI will fix this. For text-based work, it largely has. But the physical world does not run on text. It runs on what people and machines can see.
For a transportation agency managing highways, an offshore operator monitoring a rig or a hospital tracking patient safety, video is often the richest signal available, and in many cases the only one that captures what actually happened.
When it comes to video, the promise of AI has outrun the delivery. Most organizations are running a video management system built for recording, not detection, and the AI features they were sold require new cameras to unlock. The turnkey platforms that promise to skip that problem lock operators into one vendor's models and one vendor's cloud. Building it in-house produces a pipeline that breaks the first time a model changes. The result is a lot of demos and not much production.
That gap matters because when video is connected to systems that can act on it, the outcomes are concrete. A wrong-way driver detected in three seconds versus three minutes is the difference between a closed ramp and a fatality, as the Federal Highway Administration estimates wrong-way crashes are 27 times more likely to be fatal than other highway incidents. An offshore operator that catches a pressure anomaly or an unauthorized entry into a red zone in real time avoids a shutdown that can run into the millions per day. A hospital that flags a patient fall in a corridor shortens response time, and falls are among the most expensive issues for inpatient care.
This information is most likely caught on video that can be reviewed after the fact, but if organizations could see and respond to real-time events, there would be huge cost-savings and harm reduction.
So why hasn't this happened already? For a long time, three things got in the way.
First, the intelligence in a video system lived in the camera itself, so detecting something new meant replacing hardware. This is fine for small deployments, but is untenable across thousands of cameras and distributed facilities.
Second, security concerns kept video locked inside the building, because for hospitals, corrections facilities and critical-infrastructure operators, footage leaving the network is a real risk, not a theoretical one.
Third, the economics of running every stream and every frame through a hyperscale cloud got difficult fast, especially once continuous AI analysis entered the picture. Hardware has gotten more flexible. The other two have not.
One of these three places is where most operators are stuck. Even so, the path forward is visible in the deployments that are already working. Consider a state transportation department managing thousands of roadway cameras installed over a decade. The cameras work. The feeds are clean. The agency now wants to detect stalled vehicles, debris in travel lanes and wrong-way drivers, none of which the original hardware was specified to do. Replacing the fleet is a multi-year capital project. Layering detection on top of the existing feeds is a software decision.
The same pattern plays out offshore. Drilling operations have spent years deploying cameras for safety and compliance, and those feeds sit on bandwidth-constrained links from rigs to shore. The opportunity is not more cameras. It is using the existing ones to flag pressure anomalies, perimeter breaches or personnel in restricted zones before a small issue becomes an incident report.
Even live sports, where the cameras are the product, run into the same constraint. Leagues want automated highlight detection, foul-play review and audience analytics from feeds they are already producing.
What ties these examples together is a shift in where the intelligence lives. The industry is moving toward a model where detection and analysis are decoupled from the hardware, and where the choice between on-prem and cloud stops being a choice. Organizations should be able to define what they need their systems to look for, run that intelligence wherever their security posture requires and change their minds next quarter without a procurement cycle.
The cameras are running. The data is there. The organizations that figure out how to act on it in real time and at scale will operate differently than the ones that don't.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。