I'd like to detect if camera view is obstructed (fully or partially) by some object (paper, hand, some device, etc).
Do you know any good methods for this task?
I've already experimented with:
RGB histogram comparison of previous and current frame (BHATTACHARYYA metric)
search for large changes (relatively big contours) on a foreground mask retrieved from MOG background subtractor
My results are not satisfactory. For first method, changes on color histogram between consecutive frames are too small. Results are better when there is greater distance between frames like comparison of n and (n-10) frame, but still this method can produce silly false positives. The second method is less resilient to light conditions. Also it may fail, because contour of a single object is not fully discovered or there are smaller, separated contours detected.
The histogram method is better, but I would like to ask your opinion about whether I approach this problem well and whether it is worth trying something different.