Machine vision

The term Computer Vision and Image Understanding generally describes the computerized solution of tasks that are geared to the abilities of the human visual system.

Especially seeing machine systems are currently used in industrial production processes in the areas of automation and quality assurance. Other areas of application are, for example, in the transport industry - from simple speed trap up to "seeing vehicle " - and in security systems (access control, automatic identification of hazardous situations).

The following tasks can currently be solved economically meaningful:

  • Defect detection under surfaces
  • Shape and dimensional inspection
  • Position detection
  • Surface Inspection
  • Object recognition
  • Layer thickness measurements
  • Completeness check

Only a comparatively small part of the current research projects is concerned with actually to understand the meaning or the content of images; most cases it is rather to detect objects in images to describe them to measure their properties to classify them, and to take on these results, decisions or controlling processes. Since it mostly comes to the design or application of computational methods in the image understanding, it is a branch of computer science that has strong relationships with cross Photogrammetry, signal processing and artificial intelligence. The tools of the machine vision generally comes from mathematics, in particular geometry, linear algebra, statistics, operations research (optimization) and functional analysis. Typical tasks of machine vision are the object detection and the measurement of the geometric structure of objects and motions ( foreign motion, self-motion ). In this case, use is made of the image processing algorithms, for example the segmentation, and to methods of pattern recognition, e.g. for the classification of objects.

Methods

Tools of image processing for automated interpretation are:

  • Farbklassifikatoren, for example for skin color classification
  • Hough transform and contrast analysis for the detection of geometrical objects
  • Optical Flow for motion extraction
  • Sobel operator, wavelets, Gaussian - Laplacian pyramid, Gabor wavelets and Laplacian Of Gaussian filter for edge detection

In more complex recognition tasks models are often used. These include knowledge that can be used for detection of an object. For example describes a face model that the nose must always be located between the mouth and the eyes. Thus, a search algorithm know roughly where to look mouth if he has already found his eyes and nose. Here are some modeling techniques:

Applications

The techniques of image understanding are nowadays successfully used in industrial environments. Computer support, for example, the quality control and measured simple objects. Largely determined by the programmers here the environmental conditions that are important for an error-free draining its algorithms ( camera position, lighting, speed of the assembly line, position of objects, etc. ).

Examples of use in industrial environments are:

  • On a conveyor belt washers are checked to verify the dimensional accuracy and the error rate of the final product to decrease by several orders of magnitude.
  • Welding robots are controlled to the correct welding position.

Far more difficult demands on the techniques of image understanding in natural environments. Here the programmer has no influence on the environmental conditions, which significantly complicates the creation of a robust, error-free running program. One can get this problem is an example for the detection of automobiles illustrate: A black car stands out against a white wall well as the contrast between a green car and a meadow, however, is very low and a distinction not easy.

Examples of use in natural environments are:

  • Automatic detection of the road and pedestrians on the roadside
  • Detection of human faces and their facial expressions
  • Recognition of persons and their activities

More applications can be found in a variety of areas:

  • Automation
  • Contactless 1D, 2D and 3D measurement ( photogrammetry) for remote sensing and quality control
  • Gesture recognition
  • Medical
  • People detection ( face recognition, facial expression recognition, biometrics)
  • Robot vision
  • Characters and character recognition (OCR, handwriting recognition )
Outline of object recognition Sobel-Operator Facial recognition system Optical character recognition Image analysis Computational visualistics Integrated Authority File
124448
de