Socionext Uses AI for People Recognition in a Snap
February 07, 2019
DeepVision?s products generally lie at the edge of the IoT, and combined with the expertise of Socionext, full-fledged SoCs can be realized.
One of my favorite “under the radar” companies is Socionext. I’m continually impressed by the company’s technology, and that was again the case at the recent CES show in Las Vegas. The company specializes in image processing products and recently partnered with DeepVision, a vision-based machine-learning vendor. DeepVision’s products generally lie at the edge of the IoT, and combined with the expertise of Socionext, full-fledged SoCs can be realized.
The demo that I was shown was a multi-face detection and recognition solution, likely intended for surveillance applications. In a matter of seconds, my face was scanned using a standard mobile phone, and the system easily recognized me among a crowd of people. It’s easy to see how the good guys could use this technology to find the bad guys. The company claims that the technology could be used for emotion recognition and driver-drowsiness detection. There are lots of jokes attached to the former, but the need for the latter in automotive applications is huge.
It works pretty well in poor lighting conditions and even with lots of other stuff going on in the background. Depending on the use case, neural network configurations can range from five to more than 30 layers, and the performance can scale up to hundreds of AI inferences per second. Power can be scaled depending on the needs of the application, going down to just 100 mW.