Object recognition
17 year

i am busy on my graduation, and a part of my project is to write a module to detect objects. The object i have to detect, is a rectangle, about 5.5cmx8.5cm, and 98% of the time it has a light color (almost white). I dont care about the other 2%. The big problem is the orientation of the object:
it can be turned 360 degrees, in almost all directions. So the object isnt always visible as a rectangle, but it can be a seen (on the image) as a rectangle without angles of 90 degrees.

I was thinking about the following:
- convert my image to greyscale
- filter all light objects
- do a houghtransform to detect all the lines
- with those lines, find a rectangle

is this the (or a) right way to do it, or do you have other hints for me?
Square Detection
17 year

Yes, I think you are on the right track. It will largely depend on what kind of speed you are looking for in developing that module. The hough line transform will work but it is quite slow. From the transform you would need to find intersecting lines, determine the corners and from that deduce the square.

This assumes that the object is amongst other objects within the image view. Quicker techniques involve segmenting the image (as you mention using thresholding), connecting all touching blobs, and then do a quick check on the area versus perimeter. If area = side * side and perimeter = (side*4) then you can solve for what the side length should be in each equation and see if the are approximately the same. This will give you a rough estimate of how square the blob is. This may be useful as a filter prior to the slower hough transform.

There are other ways to detect lines ... we're working on one of those technique built specifically to detect rectangular objects at a rapid pace but the technique is just a fast implementation similar to what I described above.


dabbe  [3 posts] 13 year
ı wanna detect 2 cubic object due to time. Let s say i wanna sense if one of the cubic object moved less than a specific time or not? I am so new about image processing plz help me.
Anonymous 13 year

This sounds possible but you should include two images that show what you are describing. I realize that English is not your first language so having an image or two would greatly help us to understand what you are trying to do.

dabbe  [3 posts] 13 year
Sorry for my English. Its true i m not a native speaker

In my project i wanna recognize 2 different objects. They are sumo robots. I mean 2 cubic objects. Their color can vary but mostly black.
dabbe  [3 posts] 13 year
And an other important issue is the ring of the sumo robots is black metal surface which is surrounded by a white line. I am trying to recognise a black object on a black surface. Like in the photo...

Anonymous 13 year
You will need to place the camera directly above the arena to get a good angle. You can then use the attached robofile to detect the circle.

We need images for your previous question. Images are always required!


This forum thread has been closed due to inactivity (more than 4 months) or number of replies (more than 50 messages). Please start a New Post and enter a new forum thread with the appropriate title.

 New Post   Forum Index