Color feature selection for smoke detection in videos

This work presents a color feature selection for white smoke detection in a day light. This scenario was chosen since this work was applied to environmental conditions. Firstly, a manual segmentation of real world images is made such a way to define the colors associated with the white smoke. A set of 1,106,340 samples was defined. Secondly, a set of 29 features created from several color models were extracted. After that, the margin samples were selected based on the margin criteria, achieving a final set of 4,860 samples. Finally, the channels of the color models were ranked by relevance for the development of smoke classification models using the Relief feature selection.