3
$\begingroup$

I need to detect the exact position of some kitchen tools using machine vision. I've the possibility to label the tools with a pattern, to make the detection (more specifically the image segmentation) easier.

What is the easiest pattern to detect using machine vision?

The pattern should be:

  • Easy detect in presence of partial occlusions.
  • Easy to detect with varying and degraded light conditions.
  • Easy to detect with widely used computer vision algorithms, in particular lightweight algorithms that can run on edge devices.
  • Hard to mistake for another object (in a kitchen).
  • Possibly carrying some additional information.

A QR code fits this description pretty nicely. My guess is that the easiest pattern will contain a sharp alternation of two colors, likely black and white.

$\endgroup$
7
3
$\begingroup$

In case you can shoot a video of the static scene than a blinking light would be the easiest as you could easily detect it by subtracting the n - 1 frame from the n frame until you see something with high values.

If you take a still shot you can use 2 main ideas:

  1. If the colors of the scene are from a given plate, find a color very different in Hue and make sure it is intense (Even active lighting which will saturated).
  2. Use a pattern which is easily detectable like stripes of B/W or pure green and magenta. You may encode data in the stripes.

Resources

Some resources pasted here in comments:

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.