Detecção de incêndios ambientais utilizando persistência espaço-temporal, segmentação por cor e sub-amostragem de vídeo

Abstract

This work proposes a method/algorithm of environmental monitoring in order to detect outbreaks of fires in environmental areas through video images, using pixel segmentation by color in the YCbCr color system. Characteristics obtained by another color system (YUV) were used here to generate a new classifier and improve the color classification performance through a committe of classifiers. To better address the problems found in outdoor environments and to extract features relevant to the problem, a background subtraction method called running average was incorporated, using colors instead of gray scale. In addition, a new model based on ant colonies was proposed to assess the spatial and temporal features of fire in order to filter objects with the color of fire, reducing the rate of false positives caused by these objects. A subsampling mechanism for different resolutions was implemented and an analysis of the time spent in processing the videos on each resolution was done to evaluate the possible use of these resolutions in real time. Experiments were carried out and results compared to relevants methods in the literature, where it is pointed out that the proposed method was able to reduce the number of false positives while maintaining a top ranking or being close to the results obtained by other methods, as well as surpassing those in the category of processing time for higher resolutions. Furthermore, it was shown that the method can be used outside of the proposed context with relative ease, like an industrial environment, for example.