- Method 1: in the directory "visibility_detection_by_contrast", the contrast method is used. The code took the 4-nearest pixels to calculate the contrast value(In the paper[1], it's said that use 2-nearest pixels can get the same effect while saving much time) -- if the contrast value of the pixel reaches the threshold and it's not a very isolated one, its y_axis value is perhaps the visual limit. The paper referred is below.
- Method 2: in the directory "visibility_detection_by_2nd_derivative_only_road_surface", referred to paper[2]the 2nd derivative method is used as:
- Get the vector containing the mean of gray values of each line in a road scene image.
- Calculate the 2nd derivative of the gray values, and find where the 2nd derivative equals 0, of which the y_axis value is probably the visual limit. Besides, I restricted the visual limits in the road region, to get rid of the interference of the trees or other things around.
- However, the candidate y may exist a lot, so the code here took the result of method 1 -- select the y which is nearest to the result in method 1.
- It seems that method 2 is better than method 1 here, while referred on results of method 1.
-
method 1:
-
method 2:
Paper1 -- 李勃,董蓉,陈启美.无需人工标记的视频对比度道路能见度检测[J].计算机辅助设计与图形学学报,2009,21(11):1575-1582.
Paper2 -- Hautière, Nicolas & Tarel, Jean-Philippe & Lavenant, Jean & Aubert, Didier. (2006). Automatic fog detection and estimation of visibility distance through use of an onboard camera. Mach. Vis. Appl.. 17. 8-20. 10.1007/s00138-005-0011-1.