Smoke detection is very interesting task for image processing. Smoke has features such as transparency, unsettled shape, colors and so on, and the conventional image processing method cannot handle those features well.
We try to understand why the smoke detection can be easily done by human and image processing cannot do as human does.
Note: If you have a problem to play movies, please check your ad block setting on your browser and PC.
SVM based smoke detection
(see "Smoke Detection in Open Areas with Texture Analysis and Support Vector Machines", IEEJ Trans 2012; 7(S1): S59–S70)
LBP based smoke detection
(A part of this work was shown in "Anisotropic LBP descriptors for robust smoke detection", Proc. IECON 2013 - 39th Annual Conference of the IEEE Industrial Electronics Society, Vienna, 2013, pp. 2372-2377. )