Spatio-temporal Smoke Clustering in Outdoor Scenes Based on Boosted Random Forests
URI (for links/citations):http://www.sciencedirect.com/science/article/pii/S1877050916320415
Институт космических и информационных технологий
Кафедра систем искусственного интеллекта
Journal Name:Procedia Computer Science
Journal Quartile in Scopus:без квартиля
Journal Quartile in Web of Science:без квартиля
Bibliographic Citation:Margarita, Favorskaya. Spatio-temporal Smoke Clustering in Outdoor Scenes Based on Boosted Random Forests [Текст] / Favorskaya Margarita, Pyataeva Anna, Popov Aleksei // Procedia Computer Science. — 2016. — Volume 96. — С. 762-771
Текст статьи не публикуется в открытом доступе в соответствии с политикой журнала.
Nowadays, vision-based techniques for automatic early smoke detection in the outdoor scenes are in a hot topic of computer vision. The basic set of features includes the traditional features describing the spatial ones, such as color, shape, transparency, energy, and fractal property, and the temporal ones, such as frame difference estimator, motion estimator, and flicker on boundaries. The main problem of the early smoke detection is to obtain the low values of the clustering errors. Our contribution deals with a reasonable clustering of the smoke/non-smoke regions based on the Boosted Random Forests (BRFs). The BRFs provide better clustering results in comparison with the traditional clustering techniques, as well as the ordinary random forests. Forty test video sequences with and without smoke were analyzed during experiments. The true recognition results of a smoke detection achieved 97.8% that is better on 3–4% of the results obtaining by the Support Vector Machine (SVM) application. False reject rate and false acceptance rate values were significantly decreased till 3.68% and 3.24% in average, respectively.