Picture this: AT&T sees connected cameras dominating IoT
by Martha DeGrasse
Each business day, AT&T moves 100 petabytes of video across its wireline, wireless and satellite networks. Most of that video is downloaded by consumer and business customers, but AT&T expects video uploads to increase dramatically as companies connect more devices to the internet of things.
"Taking a high-resolution camera and attaching some very sophisticated image recognition software to it, you can do some amazing things," said AT&T CTO Andre Fuetsch. Fuetsch sees image recognition as one of the most powerful artificial intelligence applications. Last fall at the AT&T Summit, he outlined several of the ways AT&T customers can use image recognition through the company's Acumos AI platform.
Fuetsch said the Acumos platform will allow developers to easily integrate separate algorithms, so one algorithm could identify images and another could attach meaning to those images. The result could be a solution that tells a retailer how often people look at an in-store promotion, how they react to the ad, and which age group or gender seemed to like the ad the most.
Image recognition software can also give retailers more information about their customers when cameras capture images of hands taking products off the shelves or presenting them at the cash register.
"You can learn a lot about your customer real-time in terms of the demographics: gender, age range, ethnicity," Fuetsch said. Of course that data would be anonymized unless a retailer tried to integrate image recognition software with a system that stores identification data.
Cameras can also be used to replace other sensors in IoT deployments. Load sensors, for example, are not needed if cameras can photograph freight and software can interpret the image. In some cases, even temperature and humidity sensors can be replaced by cameras that photograph drops of condensation on a product, paired with software that measures the size of those drops. Temperature and humidity can be derived from the size of the condensation beads on an aluminum can, Fuetsch said.
Public safety is another important use case for artificial intelligence and video analytics. Body cameras worn by police officers can use algorithms to decide when to record or stream video.
"There's this whole issue around when you turn the body cam on, when you turn it off. ... Policies can be built so that you don't leave it to that junior rookie officer to make that decision," Fuetsch said. "The policies dictate when the cameras can turn on and off." He isn't talking about police force policies - he means network policies enabled by artificial intelligence and video analytics.
"We are a video dominant network and that causes us to design and engineer to be very video-centric," Fuetsch said. "Of course most of that video today is downstream, but upstream is going to become, in my view, an even larger force. ... If you think about what is the killer IoT app, it's the camera."
Follow me on Twitter.
