I am a total newbie, asking the universe:
has anybody worked with roboflow to detect changes between two images?
I would like to use this to find changes in a long jump sand pit…“where did the jumper land”?
Is it possible to train this kind of detection, when there is no actual “change” in the appearance, only in comparison…?
Hi @AlexanderH
Absolutely! The applications of object detection and computer vision in general are limitless.
Once you train a model that can detect the location of your objects (jumpers in this case) and then from there, you can create anything you’d like to modify the results from there.
I think this resource might be helpful for you: Measure Distance in Photos and Videos Using Computer Vision
Thank you for the interesting link 
The distance estimation is “under control” in another program , KINOVEA 9.6. The program uses a calibrated grid on top of the image and lens correction to reach an accuracy of <0,5 cm in measurement when filming from a very slanted perspective. The problem is the identification of disturbances/changes in the sand surface before&after the jump. The objects are only changes (shadows) on an unevenly, brownish background. They have no discernible shapes or definite changes in comparison with the background…
Hi @AlexanderH
I see. I think that could be a use case for computer vision here, but I think it might be better to see if it is possible to detect more discernable changes if you wanted to use CV.
As others have pointed out, I think this could be a problem for more classic CV algorithms to be applied to, perhaps augmented with ML.
A possible attack would be via the Canny Edge detector and/or Hough Lines transform, maybe you could apply a template ?
Does the ground get raked/smoothed between contestants ?