0

Suppose we have a device with a camera. We would use device camera and optical flow to detect the device 2D displacement. To do this we need to examine a stationary background framed by the camera.

It generally doesn't work when the background is a low contrast/saturation image. For example: sky, asphalt, floor, white surface (ceiling)

Is there a way with digital image processing or some algorithm to make optical flow usable in these cases?

Adriano Foschi
  • 648
  • 1
  • 8
  • 23
  • 1
    Low contrast..... perform contrast enhancement !! – Jeru Luke Feb 07 '17 at 16:36
  • I did not understand the part where you say _when the background is a low contrast/saturation_ and then you give examples as sky, asphalt etc. Opencv has [examples](http://docs.opencv.org/3.2.0/d7/d8b/tutorial_py_lucas_kanade.html) of Optical Flow on asphalt roads. – NAmorim Feb 07 '17 at 16:37
  • Imagine that your smartphone camera is framing a white wall. You would see some light shadows. You have not a considerable point of contrast that you can choose as the point to track the displacement. The question is how to choose "the best" point to track the displacement in these conditions? – Adriano Foschi Feb 07 '17 at 16:42
  • consider viewing [THIS POST](http://stackoverflow.com/questions/24341114/simple-illumination-correction-in-images-opencv-c) for an initial insight – Jeru Luke Feb 07 '17 at 17:03
  • @Adriano Foschi I believe that what you described is the very weakness of **pure** image based tracking. – NAmorim Feb 08 '17 at 12:41
  • For optical flow estimation you don not "need to examine a stationary background" what you are describing are background substraction or modelling methods. Optical flow approaches have some problems with homogenous areas but recent approaches have overcome this limitations – Tobias Senst Mar 27 '17 at 18:31

0 Answers0