corner detection accurately

asked 2015-06-25 10:25:54 -0600

Suman_83 gravatar image

updated 2015-06-25 10:27:49 -0600

I need help in my current project in which i am stuck up in one of the module.I have image from which i am required to get the dimensions of the object that too accurately as from these dimensions i would have to point out the faults in the structure of the object.

These are the problems that i am looking solution for

1.i am required to get the corners of the object in image that i have attached.( To make it clear i have rounded the corners that i want).

I have tried out harris corner algorithm,Approx poly and few other solutions that i could find out in the net but none of the methods brought me close to the required result (few things that i have tried out "

" )"

2.I am having a turntable type setup for capturing the object image using single camera.i am looking for suggestion for the best possible way to calibrate the camera accurately as my current setup is small which is placed in 30mmx30mm box

3.i have structure of the ideal object , i have to compare that to various other objects and accurately point out the faults in the dimensions and angles of the examined object compared to the ideal object.

Please do excuse me for asking all my problems in one place but all of them are interrelated so i had to put it together

I am new to this so please do help me.I am looking for guidance and ways to proceed as i have gone out of ideas. sample image

edit retag flag offensive close merge delete


I have never done that but my idea is first to look for a contour of this shape.

Second try to compress it. Compression mean for one point will have length and angle to go the next point.

Compression algorithm is just calculated vector between point i and i+1 and if this vector is equal to previous point there is no need of new point and increase distance and go on. Of course if your shape is noisy previous method is not very good

For calibration you can use this sample

LBerger gravatar imageLBerger ( 2015-06-25 14:51:49 -0600 )edit

Thank you for the fast response and also the suggestions for the -corners i will be trying to implement the compression algorithm suggested by you it might work out to be the way but as pointed out by you i have to check for the noise -coming to camera calibration the chess board example is good but there is a limitation in my environment as i pointed out in my problem, the entire setup of the camera and the object is small (30mmx30mm box) so how can i use chess board in this small setup or is that i am missing out something here

thank you once again for helping me move forward in the project

Suman_83 gravatar imageSuman_83 ( 2015-06-26 02:46:53 -0600 )edit

About calibration may be you can use 3D printers to make yourself or make an electronic card. With a chessboard make with a printer I don't know what is printer accuracy. It could be a problem

LBerger gravatar imageLBerger ( 2015-06-26 06:16:23 -0600 )edit