Intel Offering Combines AI and Vision
May 16, 2018
With Intel CPUs, FPGAs, and VPUs, deep/machine-learning gets a huge boost.
AI and vision systems. That sounds like the perfect combination. That’s because it is, and now it’s part of the Intel portfolio. Intel recently announced the OpenVINO (Open Visual Inference and Neural Network Optimization) toolkit, which lets developers combine a graphics subsystem with the algorithms required to deployed a sophisticated AI/machine-learning platform at the Edge of the IoT.
The application areas for such a solution are too numerous to name, and many more will arise as the technology proliferates. Off the top of my head, I can see how it would expand the shopping experience. Walking in front of a camera lets the system know your weight and height, your age and gender, just to name a few characteristics. It can also infer from how you’re dressed or who you are shopping with what your tastes are. Then the store either highlights certain areas or works through your mobile device to enhance your shopping journey.
A second obvious situation is for traffic management. In addition to simply observing the traffic flow, the system can consider weather (current and future), time of day, holidays, unplanned events, and so on. With all that information, things like lane opening and closings, parking spot availability, traffic-light patterns, etc., can be adjusted accordingly.
So what makes up OpenVINO? It consists of Intel CPUs with integrated graphics; Intel (formerly Altera) FPGAs; and the Intel Movidius video-processing unit (VPU). Don’t underestimate the concept of having a programmable component in the equation. That allows the designer to adapt to almost any potential network or interface. And the VPU, which obviously specializes in video processing, can help maintain power and performance efficiency.
This combination also excels at making a pair of key decisions—what and where to send to the Cloud verses handing at the Edge. This is important for applications that need to operate in real-time. While sending information to the Cloud for processing can be quick and efficient, it’ll never be confused with “real time.” In addition, depending on your transmission medium, transmitting data to and from the Cloud can be an expensive endeavor.
As a validation point, Intel has announced that it is already working on end products with such partners as Amazon Web Services, Dell, Honeywell, and multiple divisions of GE.