Detecção de fumaça em vídeos para monitoramento de áreas ambientais

Detalhes bibliográficos
Ano de defesa: 2013
Autor(a) principal: Adriano Lages dos Santos
Orientador(a): Não Informado pela instituição
Banca de defesa: Não Informado pela instituição
Tipo de documento: Dissertação
Tipo de acesso: Acesso aberto
Idioma: por
Instituição de defesa: Universidade Federal de Minas Gerais
UFMG
Programa de Pós-Graduação: Não Informado pela instituição
Departamento: Não Informado pela instituição
País: Não Informado pela instituição
Palavras-chave em Português:
Link de acesso: http://hdl.handle.net/1843/BUOS-9MNKTB
Resumo: Fires are one of the main causes of deforestation, destroying a large percentage of the woods and forests. Moreover, they are responsible for producing and emiting a great amount of CO2 into the atmosphere and, on the top of forest destruction, it also causes ambient unbalance. An efficient solution to reduce and to prevent the damages caused by forest fires, is to detect and extinguish fires as fast as possible, avoiding them to evolve quickly for an uncontrolled fire and of great proportions. One of the forms to reach this objective is to create systems that anticipate fire detection, in this case detecting smoke in the beginning period of the fire. This work aims to develop a smoke detection system by means of video sequences, where the purpose of the system is the smoke detection in early stages of formation. The analysis of the points of interest candidates to smoke consists of detecting objects in movement in the videos, through background subtraction. After detection of an object in movement, it is classified according to its color. Only objects with smoke color are analyzed by the last stage, that carries out analysis of space movement, besides verifying the temporal persistence of pixels of interest in the frames of the videos. Pixels that satisfy rules defined in the three stages are considered pixels of smoke. The system considered in this work is compared with other existing systems in literature. Through tests with a database with videos that contain smoke images and videos that do not contain smoke images, but contain objects that look like smoke. The accuracy of the proposed method obtained in the tests was 8%-30% over existing methods in the literature which relies on chromaticity analysis, motion detection and signal processing. The accuracy of the proposed method was confirmed by statistical analysis.