Introduction to machine vision

How machines and robots see? This post’s subject is an introduction to machine vision. It is very used in industrial machines and robots.

The system

Every system with machine vision captures a image, makes an analysis with software and answer in an adequate way. The most important components to a machine vision system are: one or more high resolution cameras, digital analog converter to image digitalization, a processer, big quantity of RAM memory and an artificial intelligence (AI) algorithm.

Improving machine vision

Exist some techniques to improve machine’s vision, increasing the efficiency of robot and process.

Illumination: Lighting up the inspected local can increase the contrast and precision of captured image. This graphic shows the relative intensity in relation to wavelength of most used lamp types in machine vision.

Trigger range function: The industrial environment can create noises, provoking fake triggers in the system, in the software must be defined what the camera and sensors must detect to obtain the proper answer.
Filtering: Some lens are used in cameras to filter a determined wavelength. To improve quality image and precision.

Image recognition

How a machine recognizes an image? All image is seen as an array with 2 or more dimensions. Each matrix’s element is represented as a pixel and a number. This is the representation of an image with 10×10 pixels.

Each number represents a pixel’s color and each color is the combination of primary colors: red, blue and green. Great variation in pixels indicates the presence of objects or figure’s features.

A convolutional neural network is used to recognize objects and figures. First is showed a set of images, each image has some features which define your classification. The network is trained to find the features and classify properly the images. Then a test is made to verify if a neural network learned to make the correct classification.