MEI-56606 Machine Vision

• Consider an application, where machine vision system’s task is to locate non-transparent parts from specified area (200 x 200 mm) and tell part coordinates to robot. At any given time robot needs to pick only one part type; however, part type does change every couple of hours. Parts can be recognized based on their shape colour information is not needed.

Robot needs to know part location with 0.75 mm accuracy so that picking is reliable.

1. What system type (PC based / smart camera) would you use? Why?
2. Colour or grayscale camera? Why?
3. What kind of illumination would you use? Why?
4. What camera resolution would you use? Why?

• We have a 10 mW green laser pointer (λ = 530 nm). What power should a red laser (λ = 650 nm) have so that both lasers seem equally bright to human eye? Assume that the laser spot sizes are equal. (Hint: Document ‘Radiometry.pdf’ (additional material for lecture 2) in Moodle might be helpful.)
• Below we have a 10 x 10 pixel 3-bit grey scale image.
  1. Calculate the histogram of the image. Give the histogram both in graphical form and as a table.
  2. Select a global threshold value and threshold the image to segment object(s) from the background. Result is a 10 x 10 binary image where you have marked pixels forming the object(s) with ‘1’ and pixels belonging to background with ‘0’.
  3. Calculate the area of the largest target in pixels. If there are more than one target in the image, calculate the area of the largest target.
  4. Calculate the mass centre point of the target and mark it into the binary image. If there are more than one target in the image, calculate the mass centre point of the largest target.

5) Calculate the depth of field for the following system: field-of-view 1200 mm wide, camera IDS µEye UI-2240-M (detector is 5.95 mm wide with 1280 pixels), and a lens with 12 mm focal length and aperture adjusted to 2 mm in diameter.

This was not discussed during lectures so a “small” hint: In his doctoral thesis Dr. Uusitalo has presented an equation for estimating the depth of focus of a sensor system (Uusitalo, Jani. Multimodal Laser Triangulation as a Tool for Industrial Three-Dimensional Reverse Engineering and Quality Control, Tampere University of Technology, Publication 600, Tampere 2006, pages 56 – 58):

2z = 2 DN(M + 1)/M²

where: 2z is the total depth of field

Dyis the pixel-to-pixel distance of the camera detector

N is the numerical aperture of the lens, and

M is the magnification of the imaging system.