Road map to solve recognition problem

I’m maintenance department worker (electric + automation) on my factory is some application which recognition wheel (model from small active element list, and measure angle). I want try wrote similar application and teach openCV library. At the moment i Test openCV image manipulation functions. But lack me global approach to the problem. The roadmap what library elements to use to solve the problem. At the moment i wrote openCV-Qt test program which do “AND” operation from two different picture(the same model). Second picture is rotated and rotation angle is calculate from mouse click position. If you have any suggestion for me. link to video with my problem is below
2022 03 03 09 36 17 - YouTube.
For any suggestions and road map proposition ,thank you


I have trouble making sense of your text. perhaps try that again in your native language and use machine translation.

I am so sorry for the confusion !!!

I want to collect suggestions on how to approach the problem of measuring the angle of rotation of the wheel (as in the movie)

at first I thought about removing the background from the image by superimposing two photos, but it s a wrong idea or a wrong method. merge with mask does not solve background problems in holes

commercial solutions exist for this. the keywords are “machine vision” (MV) and “automated optical inspection”.

to break this down:

  • locate wheel
    – separate wheel from background (maybe)
    – find circles, center
  • relative to wheel, find valve stem (?)

based on the video…


  • maybe houghCircles will find the hub and rim circles
  • MV approaches would sample the picture along several radial profile lines, arranged like a circle were to be found at a certain position, find edges in those 1-dimensional profile lines, and use those points to calculate the actual circle. that’s a lot faster than Hough, a lot more precise, and easier to debug.
  • image may need more operations to distinguish the wheel from background

valve stem:

  • MV would take a circular profile line at a radius where the valve stem is to be expected (relative to the real circle), then analyze that 1D signal for distinguishable edges/levels
  • the equivalent in OpenCV would be warpPolar

you might want to google for some commercial MV programs and look at some screenshots. a few illustrative examples:

thank you for your comments, I’m starting to learn