Computer vision on low cost and low power hardware? PoC on Arduino Portenta H7
One of the challenges AISent is working on is bring the computation “at the edge”, distributing intelligence to devices able to run inference (i.e. the predictions) and possibly the training without relying on powerful machines. The less the power, the greater the challenge.
We decided to test us experimenting the Arduino Portenta H7, a low cost board based on the STM32 Arm processor and its Vision Shield integrating a 320 x 320 pixels camera (0.1 MPixel). The sum of both the boards is around 100 euros.
The Proof of Concept wanted to demonstrate the possibility of doing inference on a low power board. We created a model able to see if a bottle was capped or not, real time using computer vision algorithms and a classification model in Tensorflow Lite.
You can observe the first result in the following video on our LinkedIn page.
A big applause goes to Arduino, STMicroelectronics and the creators of the OpenMV project that really helped us prototyping the system. Looking ahead for the industrialization, some consideration we could do are
Bringing AI to the edge devices is a wide topic that covers efficient implementations, models and algorithms designed to perform at their best on the selected hardware (going from a micro-controller to an industrial PC), robustness and trustworthiness of the results, etc.
To continue following the updates follow us on LinkedIn or here on the blog.