Machine vision

Machine vision

  • Olivier Algoet
    • Olivier Algoet
    • 07/10/2021
  • In recent years, we have become increasingly convinced that there is little that machines cannot do for us. Nevertheless, there are actions humans do every day that pose a considerable challenge to a machine. Case in point: the visual recognition of objects. Triggered by this challenge, Squadron member Olivier developed a proof-of-concept program that allows machines to look into our world.

    Machine vision is a science in which machines can convert the real world into structural data. Machines can thus more easily identify people in a photo or in real life but can also drive a car because they recognise obstacles. But why is this such a challenge for machines? Quite simply, life is sketched by small nuances. Think of lighting, a different angle, people sitting or standing in an unusual position… The list goes on! Every object is different, and a person certainly is! Recognising emotions can also be a real challenge for both man and machine.

    Yet Olivier saw this as a challenge that he did not want to shy away from. He quickly started working on it in his spare time. Currently, there are few general machine vision applications that can combine all of the following applications. However, there are already some applications that are very good in one aspect, for example face recognition, or even a combination of some of the applications.

  • What applications can be developed with machine vision?

    • Optical Character Recognition (OCR)

    The system can identify optical characters and reconstruct them into a typed text. Think of when you are abroad and want to read a menu. Unfortunately, you do not understand the language. What do you do? You take Google and use the translation option which, via the camera aimed at the menu, shows you instant translations of what is on the menu. Here, Google cleverly uses OCR. Characters from an image are deciphered, recognised and translated via pattern recognition.

    • Object recognition

    The software can easily classify objects that are partially visible or even deformed.

  • Object recognition
    • Quality inspection

    During a quality inspection, software can quickly detect deviations and irregularities and remove them from the production line. This way, the quality of the end product can be guaranteed and the error margin is reduced.

  • Quality inspection
    • Object pose estimation

    This function identifies the position of products in 6 dimensions. This includes the location and orientation of the product. You can use it to determine whether a can has its label facing forward.

    In production environments, it often happens that goods are delivered in a structured way so that they can be easily programmed for automation. For series production, this is not a problem, but when producing high volume low batch, i.e., many different variants each time in small quantities, it can be useful to recognise the different products and still be able to work automatically

  • Object pose recognition
    • Barcode & QR Scanning

    In 2020, QR codes have finally broken through and become mainstream. When you're in a bar and order your drinks via a QR code, you're actually using machine vision. The camera of your smartphone registers the barcode and distinguishes a pattern in it. Often, these link to a website or there can be information hidden in the code that the camera translates for the user.

  • Blogpost image
  • Proof of concept

  • The machine vision that Olivier developed is a proof of concept that can identify objects and faces and recognise and name emotions in them. The software was of course tested within the community and proved to be able to correctly name all faces of the Squadron members and their emotions. How did Olivier do it?

    First, arrays were made of all Squadron members. Each face was given 52 facial characteristics, as each individual has a unique combination. When a member looks into the camera, the software will quickly perform general face recognition based on standard elements such as lips, eyes, and nose. The software then links the data it receives with the data that was entered and matches the right person. Only when the match is above a certain margin of error is the face identified as belonging to the right person. Thanks to artificial intelligence emotions are allocated. This works via a trained neural network that, based on many examples, teaches itself to recognise and name emotions.

    The applications are still in their infancy but at Squadron we strongly believe in the commercial application of this software. Because of such developments, Squadron invests in its members, so they dispose of all necessary expertise to guide a performant solution and implementation for our customers. Such applications will increasingly find their way into the manufacturing industry and will be indispensable within a few decades. This technology will conquer the world and will be part of most machines.

    Start digitising your business processes with the help of Eminds.

  • Want to stay informed?

Show all articles