|
First steps Image analyisis
Scelle from Belgium [2 posts] |
15 year
|
Hi,
I'm new to the world of vision software. I want to examine pictures with the following result: I want to detect the screw head that is about in the middle of the picture and calculate it's offset to the real center of the picture.
I came to some result but not really succesfull.
I mostly tried combining Canny edge detection followed by the Circles module.
I tried with completely different pictures and there I had better results.
Does anyone have any idea what I would be doing wrong? How important is the resolution of the pictures? The light?
Am I using the wrong program or should this be no problem for a powerfull program like RR?
Thanks
kind regards
Scelle
|
|
|
|
Anonymous |
15 year
|
Scelle,
Lighting is probably the most important. You'll also need to take more images in order to determine what technique would work best. We chose to use intensity to just separate the screw from the rest of the image.
Can you post images that do not have that grid superimposed ontop?
At any rate these are the final images that we got using the max filter to remove the grid, auto-threshold to id the lower intensity parts, then using the close morphology filter to close the circle, fill to make the screw head solid and the blob filter to remove blobs that are not circular. Seems to work on the provided two images ... but most likely will need tweaking for other examples.
STeven.
 program.robo
|
|
|
|
Scelle from Belgium [2 posts] |
15 year
|
Hi,
Thanks for the quick reaction. Meanwhile I took some new pictures, closer to the screws, with the webcam leds always on and without the grid.
Now it seems to be much easier to get good results. I tried some things with canny edge detection and circles module with succes!
Concerning the light, I suppose the best option is to always use the webcam light at maximum intensity to reduce the influence of background lights.
Apparently, lots of techniques possible. I'll try to refine some things further.
Thanks for the help
scelle
program.robo
|
|
|
|
Anonymous |
15 year
|
Scelle,
That's good news! Yes, it is normally better to work with good images than with blurry ones. Your new images are much better and have a lot of nice features that can be used to detect the middle screw. Seems that the circle detection is a good way to go given these images.
In terms of lighting, it will depend. Sometimes increasing the cameras lighting will also increase the integration time that the camera needs to generate an image which slows down the framerate and increases the motion blur. Some have found that using the color_balance module will help to resolve low light images that are high in frame rate ... it really depends on your application and what works best for you.
Keep in mind that if you are looking for dark objects increasing the lighting will improve that objects details, likewise the reverse is also true.
Keep us informed on how things go. Interesting images? Are they from a packing plant? Please only answer if you are allowed to reveal a little more about your application.
Thanks,
STeven.
|
|
This Thread
