AI platforms shrink at the edge

Foghorn runs video analytics on Jetson Nano. 40Geo demos TensorFlow computer vison on Raspberry PI. OpenMV, DIY machine vision for the Arduino.

Foghorn runs video analytics on Jetson Nano

Ramya Ravichandar (Foghorn), speaking at the 2019 Nvidia developer meet in San Francisco, presented on some video analytics use cases in the industrial internet of things. Foghorn claims to have the ‘World’s smallest and fastest inference engine’ Its EdgeAI platform comes with out-of-the-box solutions. Edge computing bests a long round trip via the cloud as AI/streaming analytics is local. Real time, millisecond reaction powers time critical actions. Foghorn Lighting edge software includes VEL complex event processing and AdgeML AI. An oil and gas use case is flare monitoring on Jetson Nano showing how deep learning models can be run on minimal hardware. Foghorn is backed, inter alia, by DellEMC, GE, Honeywell, Saudi Aramco Energy Ventures and Yokogawa. More from Foghorn.

40Geo demos TensorFlow computer vison on Raspberry PI

In a different context we saw a similar AI-in-a-box on display at the 2019 Esri EU Petroleum User Group (full report in our next issue). Keith Fraley was demonstrating both 40GEO’s Raptor geo-located internet of things technology and his ‘maker’ skills. A Raspberry PI running a tensor Flow model and video camera were capable of identifying a range of objects held in front of the lens. It’s not 100%. The system thought my clementine was a ping pong ball ... a good try. What was interesting is that, according to Frayley, the compute power needed to run computer vision in real time on a small hardware footprint is not all that great.

OpenMV, DIY machine vision for the Arduino

Our own googling also located the OpenMV an Arduino-based, Python-powered machine vision bundle for makers and hobbyists. Checkout the neat IDE.

Click here to comment on this article

Click here to view this article in context on a desktop

© Oil IT Journal - all rights reserved.