Uma técnica de subtração de fundo em vídeos de monitoramento

• Ano de defesa: 2012
• Autor(a) principal: Samatelo, Jorge Leonid Aching
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Doutorado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Processamento de imagens; Visão por computador; Sistemas de reconhecimento de padrões; Vigilância eletrônica; Câmaras de vídeo; Engenharia Elétrica; 621.3
• Link de acesso: http://repositorio.ufes.br/handle/10/4122

Video surveillance applications, such as activity monitoring, tracking and identification of objects, etc, require as an initial step the detection of the moving objects existing in a video sequence. The detection step is a challenging task, because of the different variables to consider, such as the nature of the video capture, images captured in indoor or outdoor environments, variations in lighting and noise, etc. The approaches that can detect moving objects from a stationary scene are called as background subtraction techniques, and their implementation requires the development of a set of steps, namely pre-processing, background modeling, change detection and postprocessing. So, here it is implemented a background subtraction technique that includes in the development of each steps after mentioned. Consequently: (a) various techniques are performed for the background modeling step, allowing to explore different solutions and how they handle the typical challenges; (b) two techniques, that working in the color level are proposed for the change detection step, based on the Euclidean distance calculated between each video frame and the model of the background, and a thresholding operation, such that the threshold is defined in relation to a statistical significance level; (c) an evaluation procedure for determining the performance of a technique with these features is proposed, based on the calculation of a performance metric that represents how accurate is the technique of background subtraction to find the moving objects over the frame in a video sequence, and can be determined for each video or the entire database. Finally, we used two databases: the first is suitable for the development and testing of a monitoring system in public space and the second is specialized in analyzing some challenges that a background subtraction technique should handle. In the end of this thesis the results obtained are presented and some concluding remarks are highlighted.