Machine Vision vs. Embedded Vision – Moving closer together

Deep learning has gained significant attention in the machine vision industry because it does not require the complex algorithm development used by traditional rule-based methods. Deep learning inference on the edge is feasible, as there are smaller deep learning models available which are practical for embedded vision applications. We will cover the deep learning workflow from data collection to training and deployment, as well as the process of transfer learning. We will present two case studies on manufacturing inspection with inference on the edge, showing that deep learning is highly suitable for such application.

The conventional rule-based machine vision inspection methods have been applied in various industrial fields nowadays, and their limitations are increasingly obvious, especially when they encounter demanding situations such as complex imaging background where the defect features are difficult to extract. That is where deep learning method distinguishes itself. However in many occasions, the complexity of a deep learning platform is higher comparing to regular machine vision systems. Take pc-base system for example, a deep learning platform requires high-performance PC plus an independent graphics card to provide the computing capacity. Moreover, the deep learning algorithms set-up approach is totally different.Hikrobot believes that technology should always progress towards a high-end yet more practical way. With such thought in mind, our machine vision team developed SC7000 Series Smart Camera —— a deep learning based all-in-one vision system. With SC7000, the construction of a powerful deep learning inspection system becomes unprecedentedly simple.

The implementation of new technology can be an important competitive edge. Neural networks, for example, allow vision tasks to be solved automatically that previously – using rule-based image processing – required great effort, were time-consuming or simply not possible not at all. Deep learning enables completely new machine vision applications. The talk will explain how users can benefit from the new technology, point out typical challenges and show how it is possible to overcome them. It will also focus on the major steps that are required to implement and execute a neural net on an edge device.

Detect and locate objects are major tasks in machine vision. Artificial intelligence allows to refine them even further.Thanks to neural networks, industrial cameras are able to handle even highly varying objects – this also applies to challenging conditions like changing light. This talk discusses how AI-based object detection, executed directly on-camera, can support you to solve machine vision tasks more efficiently.

With an increasing number of embedded systems becoming smart(er), new possibilities and business opportunities are opening up within the embedded market. Most of these new smart embedded solutions and devices rely heavily on machine vision software to take a decision based on information contained within the acquired image data.MVTec experts will outline the specifics of the embedded vision market, highlighting the key financial and technical aspects which need to be taken into account and will try to shed some light onto the underestimated complexity that hides behind choosing the “right” machine vision software to be integrated into smart embedded devices.

GPU-powered deep learning technology is growing at an immense rate and can be found all around us. This Neousys Technology presentation focuses on how, by implementing deep learning technology can benefit machine vision applications and why Neousys’ high-powered GPU solutions thrive in extreme conditions to accelerate tasks. To see deep learning machine vision in action, stay tuned until the end for success stories.

AI can solve many tasks of visual inspection where traditional algorithmic ways reached their limits and do not work.But how does the situation look at tasks where traditional algorithms work? Is it better to solve them with AI or not? What are the benefits of using AI in these cases? Are there any drawbacks? Why use AI even if we have a solution based on classical algorithms?

Machine learning isn’t new, but the interest in it has exploded in recent years. But, do youneed to employ a bright young thing straight out of university to deploy it? Not really. It hasrapidly become a mainstream part of the machine vision toolkit and while it is necessaryto learn enough to apply machine learning, you don’t need to be a ‘data scientist’ or even agood programmer!This talk explores some of the different types of machine learning, how to get started,how to test, improve and deploy systems. The types of applications that are good formachine learning is discussed along with the pitfalls and unique considerations.