Accurate abandoned object detection (AOD) in surveillance video

• Ano de defesa: 2015
• Autor(a) principal: Alex Lopes Pereira
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Instituto Tecnológico de Aeronáutica
• Palavras-chave em Português: Sinais de vídeo; Rastreamento (posição); Indicadores de alvos móveis; Vigilância; Processamento de sinais; Telecomunicações; Computação; Engenharia eletrônica
• Link de acesso: http://www.bd.bibl.ita.br/tde_busca/arquivo.php?codArquivo=3230

The aim of the present research is to investigate the problem known as Abandoned Object Detection (AOD) in surveillance videos, where stationary objects must be classified as either abandoned or removed. We found four categories of methods to solve the AOD problem, namely, region growing, edge detection, color comparison and image inpainting and investigated all of them. We found the major drawback of each category, from which we derived guidelines that oriented the development of three novel methods. Among these three methods, the proposed method (based on edge detection) measures the ratio of a blob boundary that is covered by edges in the reference image. This method achieved convincing results on 66 test scenarios compared to results of most recent works in the literature, equaling or surpassing these results. Furthermore, we proposed a frame window scheme to combine the information from several frames about a given object in order to provide a robust classification. The experiments show our method achieved and surpassed the state-of-the-art results.