At the
Thing is,
It's clear that both Intel and Qualcomm are expecting big things soon in the field of computer vision -- the term for how computers process and understand images. The technology is becoming the latest battle filed between the two chip giants as both seek to push their computer vision technology into phones, robots, drones and anything else they can think of.
Intel has been trying to peddle its depth sensing technology to PC makers for years now -- it was previously called Perceptual Computing before it rebranded it as RealSense. It provides PC makers with a software development kit and a 3D camera to start giving PCs "senses." With RealSense, the PC camera can map out the world with highly detailed 3D scans and pick up things like depth.
The chipmaking giant has managed to get RealSense into a few laptops and at least one tablet. And now, with the recent release of
But so far, RealSense hasn't set the PC world on fire and it's not hard to see why. In a PC, all this cool depth sensing and 3D mapping technology is sitting in a constrained environment on a user's desk in an office somewhere. “The PC just isn’t an interesting platform now,” said Jim McGregor, founder and principal analyst at Tirias Research. “It’s hard to redefine the PC for new applications.”
Computer vision could really shine in devices that move around more freely in the world, such as phones and drones. “There’s just a lot more use cases for depth sensing when it’s in a portable handheld device,” said Qualcomm senior vice president Raj Talluri. “And there’s so many more phones than PCs now.”
As one of the preeminent silicon providers for the mobile age, that's where Qualcomm has a huge leg up on Intel. What Intel offers in computer vision is less about the hardware and more about the software, McGregor said. For the most part, Intel is trying to pair its RealSense software with its general-purpose CPU chips, while Qualcomm has dedicated silicon for processing images -- digital signal processors and image signal processors. That makes Qualcomm's chips much more efficient for computer vision.
“To do image processing, you need extremely low-power and high-performance chips, which is what we do in phones," said Talluri. "These are not PC processors.”
In a meeting last week, Talluri showed off a Snapdragon development board with all the components necessary for building a drone that could sense the world around it. Right now, drones are typically made from a hodgepodge of multiple vendors providing components like the processor, camera, motion control, video encoder, WiFi, GPS, etc. Qualcomm wants to sell all those functions on one board with its integrated Snapdragon system on a chip. That will make it easier and cheaper to build drones. It will also allow drone makers to start building computer vision into their flying machines, so they can start piloting themselves around and avoid flying into objects.
Still Intel has made some progress in pushing RealSense into drones. At Intel's 2015 Consumer Electronics Show keynote in January, Krzanich announced Intel was working with German drone maker Ascending Technologies on collision avoidance technology for autonomous drones. Intel then showed off a flurry of drones equipped with the Intel cameras buzzing around on stage and avoiding colliding with each other.
Besides phones and drones, the most exciting area for computer vision is in cars, said analyst McGregor. As cars start becoming more autonomous, they'll need to be able to process images and other sensor data to figure out what's around them -- be it a dog, a person, a truck or anything else.
In 2014, Qualcomm announced an automotive-grade Snapdragon chip called the 602A. It was mostly designed for infotainment systems, but computer vision is clearly on the roadmap for Qualcomm's ambitions in cars. Snapdragon-enabled cars should be coming onto the market sometime next year.
Intel and Qualcomm are hardly the only technology companies working on computer vision. It's becoming a huge area for scores of tech companies -- from chips makers to sensors manufacturers and software firms. Nvidia has been trying to apply the deep learning capabilities of its graphical processing units into image recognition in autonomous vehicles. On Monday, Microsoft announced a new research project that's trying to turn the 2D camera on the back of everyone's phones into a 3D scanner. The Microsoft researchers are hoping to come out with apps for iOS, Android and Windows phones soon.
But so-called computer vision isn't exactly a new phenomenon, pointed out Talluri. When taking a panoramic shot on our phones, for example, the camera needs to match images together and stabilize the motion. That's the same thing a self-flying drone does when it uses a camera on the bottom to stabilize itself. And pretty much every digital camera out there has some sort of computer vision capabilities inside of it -- the camera needs to understand what the lighting is like, what should the exposure be set at and where to focus. It's inferring conclusions about the world with what the camera picks up.
The difference today is that the technology is advancing and companies are finding more places to make use of computer vision. "These guys are looking for new markets," said McGregor.