Open eVision at the Edge
This paper explains the benefits of edge computing on embedded devices such as smart cameras for AI inferences, how to train and deploy a machine learning model with Open eVision, and provides practical applications examples.
Embedded computing and AI image processing are shaping the next generation of machine vision systems. Most industrial applications will call for running the Machine Vision Software inferences at the edge rather than on a remote server for latency and reliability reasons. Euresys’ Open eVision offers practical tools to train and deploy machine learning models on edge devices at an affordable cost thanks to its flexible licensing model. It can also combine rule-based image processing and AI models in one workflow for faster application development.
Why run inferences at the edge in machine vision applications?
Edge computing consists in processing image data as close as possible to its source – the sensor, as opposed to transmitting the image data to a remote computer or server or even a cloud platform for processing. In machine vision, the earliest place to process images is within the camera itself in so-called smart cameras or similar image processing devices.
- Low latency
Most industrial machine vision applications perform 100% inline part inspection. This requires highspeed image acquisition and processing for real-time decision-making to ensure rejection of defective parts. Edge computing minimizes the latency of data transmission and processing on a remote host and allows for near real-time reaction. - Security
Processing image data directly on the production line also reduces the risks related to network breakdown as well as potential hacks. Edge computing executes the quality assurance functions directly on the factory floor for more security and reliability. - Simplicity
A specific benefit of smart cameras and embedded devices is that they are very compact and can easily be integrated into industrial machines with space constraints.
Training and deploying a Machine Vision Software model with Open eVision’s free Deep Learning Studio
Machine learning opens new possibilities to vision system developers to quickly design advanced machine vision applications. Euresys has added a powerful toolbox to its Open eVision software library for that purpose: Deep Learning Studio.
Deep Learning Studio is a powerful application to train a model for a given inspection task, test it, and deploy it on edge devices. This software suite assists the user through all the steps of the development and deployment workflow. Deep Learning Studio is free of charge, so developers can train and validate their models without limitations and only pay once they deploy inferences on target devices on the factory floor.
As model training requires much more computing power than running the inference, we recommend performing the training on a Windows or Linux PC with 64-bit processor architecture, GPU (NVidia RTX 30 series with 8GB of RAM) and a minimum of 8GB RAM and 400 MB free hard disk space. Deep Learning Studio allows you to create the training dataset, annotate, augment and split it, train the model, test it, and export it for deployment.