Detect, stitch and get the contour of moving image real-time

asked 2018-10-19 17:59:58 -0500

UNOAlterEgo gravatar image

updated 2018-10-19 18:49:02 -0500

berak gravatar image

Hi, I'm working on a Python OpenCV project where I'm trying to detect a paperboard given a CAD reference file. Before using CAD, I'm using this image as a reference object.


Basically I have a cardboard object that always moves in the same direction, on a straight line, where * only a portion of the object is visible from time to time *, never the whole cardboard.

The cameras would be positioned above the cardboard, perpendiculary to the moving direction.

In order to detect I tried to use Keypoint Matching with SIFT, but not enough good key points were found. I wanted to try to use the color of the object to detect it, but I would prefer to use an approach that is independent of color, because I would like to detect different colored objects.

For the same reason, not enough keypoints, I could not stitch the different frames with the moving parts of the object using OpenCV Stitcher.

Could someone suggest an approach to solve the problem? The final result I would like to have is the precise outline of the whole object in order to compare it with the CAD file.

edit retag flag offensive close merge delete


main problem: your camera is too close to the conveyour belt, so it can't see your object as a whole. change that, not the outcome of it.

berak gravatar imageberak ( 2018-10-19 19:14:03 -0500 )edit

The image is of the whole object, in the real case, the cardboard would be much bigger like 70-80cm, and it would be visible when moving in a small gap, 20-30 cm, between two machines. Another problem is that there is no uniform background like a conveyour belt.

UNOAlterEgo gravatar imageUNOAlterEgo ( 2018-10-19 19:22:31 -0500 )edit