Detecting the volleyball net

asked 2016-07-04 02:18:08 -0600

What would be your approach to detecting the volleyball net on this images.
I am a struggling with coming up with a decent solution.
Unfortunately, bilateralFilter or gaussianBlur all result in a lot of noise (especially the trees and sand) for the canny line detector.

Can you suggest a better course of action ? Any ideas will be appreciated. Specifically, how to filter out sand and trees.

P.S. Also tried inRange based on hue, but the net and the human skin fall within the range of the sand.

Pic1:

enter image description here

Pic2:

enter image description here

edit retag flag offensive close merge delete

Comments

Will the camera always be on these positions?

can gravatar imagecan ( 2016-07-04 02:53:23 -0600 )edit

Nope. On different images camera may be located in different places. The most reasonable assumption to make about position is that the majority of the lower half of the picture is mostly sand and the net is somewhere towards the middle.

alexxxxxxx gravatar imagealexxxxxxx ( 2016-07-04 03:37:28 -0600 )edit

Well,then your best bet seems like training your own object detector.

can gravatar imagecan ( 2016-07-04 05:53:08 -0600 )edit
1

Don't use canny but rather use Sobel and start by looking only at horizontal edges, by not combining x and y directions, but rather ignoring all pixels with also a noticable gradient in x direction (vertical gradients). That would filter out the image already. Then do some sort of line detection with Hough for example, again filtering only horizontal lines. Then filter on line lengths. However keep in mind that in computer vision, garbage in means garbage out, so pictures taken against the sun will never get a decent result...

StevenPuttemans gravatar imageStevenPuttemans ( 2016-07-05 04:27:31 -0600 )edit

@can I guess object detectors is such an overkill for such a rigid object :D

StevenPuttemans gravatar imageStevenPuttemans ( 2016-07-05 04:28:11 -0600 )edit
1

Try looking at this tutorial. Get just the horizontal and vertical lines, then search for rectangles.

Tetragramm gravatar imageTetragramm ( 2016-07-05 18:20:01 -0600 )edit