repeatability precision for circular detection

[leq_82]

Hello,everyone!

First, let's make a statement ,please don't laugh at me for my poor english

OK,Let's get to the point.I need to measure the radius for the object's rounded corner(qurant),at first, I use "IMAQ Find Circular Edge 3.vi" to measure the radius,I test 100 times with statistical repeatability precision(object and it's position remain unchanged) ,I found the test results varied greatly.I adjust "Kernel Size"、"Projection width"、"Minimum Edge Strength" ......,I found the test results also bad.Then I use "IMAQ Detect Circles.vi" to measure the radius,the repeatability precision still not well.Finally,I don't measure the object's rounded corner(qurant),I measure the semi-circle and full circle on the object,I found the radius's repeatability precision is very stable.Whatever using "IMAQ Find Circular Edge 3.vi" or "IMAQ Detect Circles.vi" .From the test results ,it seems the algorithm with NI Vision find circles limited by the length of arc.

However I need measure the qurant's radius and repeatability precision is required  ,Does anyone can give me some idea?

[Hatef.fouladi]

before start to do such process you need to do some preprocessing like make image binary image filling hole or something like that also these algorithms that try to find circles always use statistical method and it couse some difference result in same image but several attempt if this difference is important for you then you can use other method like line scan of circle and find most long difference of edge in these lines to find radius