Tech Portal

Tech Portal

HomeTech PortalIntroduction to Machine Vision

Introduction to Machine Vision

What is Machine Vision?

Machine vision can be defined as a collection of hardware and software technologies that equip machines with image acquisition capabilities and allow them to automate their decision-making. In simple words, machine vision refers to the replacement of human vision and intelligence using lighting, lenses, cameras and computers.

While human vision is good at analyzing a scene qualitatively, machine vision excels at the quantitative aspect. The ability to capture and quantify a scene makes machine vision a suitable alternative to human vision for applications that require:

  • Inspection of small details
  • Non-physical contact
  • Better Safety
  • Higher productivity
  • Improved accuracy
  • Automation of repetitive tasks
  • Operation in hazardous environments

Compared to vision systems that are deployed in consumer applications for example smartphone cameras and point & shoot cameras, machine vision systems are:

  • Very high-speed (FPS)
  • Component configurable
  • API Programmable
  • Extremely robust
  • Mechanically reliable
  • And stable in extreme temperatures

Over the past decade, machine vision systems have seen explosive growth not only in terms of their capability and complexity but also in terms of their wider adoption in manufacturing and non-manufacturing applications. According to global tech market advisory firm ABI Research, total shipments for machine vision systems will reach 16.9 million by 2025, creating an installed base of 94 million machine vision systems in industrial manufacturing.

Building Blocks of a Machine Vision System

A system that leverages machine vision technologies is primarily composed of four main blocks:

  1. Image Acquisition – At the core of machine vision lies the ability to visually sense a scene and convert it into a digital format. Image sensors in conjunction with lenses can capture light, convert photons to electrons and output a digital image. This process of turning a scene into a digital image is often referred to as image acquisition. Image sensors and supporting electronics are usually housed inside a protective case which we call a camera.
  1. Data Delivery – Once an image has been acquired by a sensor and packaged in a digital format, which is called a “Pixel Format”, it is delivered to an external computing device for further processing. Here is a list of a few standards that have been developed by the machine vision industry for data delivery:
    • Camera Link
    • CoaXpress
    • Gige Vision
    • USB3 Vision
    • MIPI
    • IIDC2
  1. Information Extraction – After a raw image has been received from a sensor by a computing device, it is pre-processed and analyzed for features like:
    • Edge detection
    • Pattern matching
    • Classification
    • Segmentation
    • Measurement
    • Parts counting
    • Object Recognition and location
    • Character recognition
    • Barcode reading
  1. Decision Making – Using the extracted information, an algorithm usually trained using AI/ML/DL* would perform decision making and send control output to a machine.
Figure-4-scaled

Machine Vision vs Computer Vision

“Machine Vision” and “Computer Vision” terms are often used when there is a discussion about images so it is important that we understand the meaning behind them and clear any misconceptions.

Machine Vision Computer Vision
The term “Machine Vision” is mainly used in the context of industrial automation The term “Computer Vision” is mainly used in the context of deep learning/artificial intelligence
Machine vision is mainly associated with pre-processing of images and rule-based feature extraction Computer Vision is mainly associated with post-processing of images and learning-based feature extraction
Machine vision always involves a camera system for capturing images and data delivery Camera systems may or may not be involved
Machine vision may or may not involve GPU processing and cloud computing Large computer vision models often require GPU processing and cloud computing
The goal of machine vision is to enable machines to automate their decision-making by visually sensing their surroundings The goal of computer vision is to train an AI model by learning features from a set of images and then make predictions

It is clear that even though both machine vision and computer vision involve the processing of images, their goals are different. Machine vision is not necessarily a subset of computer vision and computer vision is not necessarily a subset of machine vision but machine vision systems often use computer vision tools to derive meaningful information for their decision-making process.