Estimativa de um Modelo Adaptativo de Cena de Fundo para Segmentação em Tempo Real de Objetos de Interesse

Márcio Oliveira SeixasAnna Helena Reali Costa

This paper describes a real-time background estimation algorithm, which has been run successfully in a bus station surveillance system. In this algorithm, a time-adaptive mixture of Gaussians is used to model the distribution of each pixel color value. For an input image, every new pixel value is checked, deciding if it matches one of the existing Gaussians based on the distance from the mean in terms of the standard deviation. The parameters of the best matching component are updated and its weight is increased. The assumption adopted here is that the values of the background pixels have a low variance and a large weight. Results are encouraging since the algorithm exhibits robustness to slow lighting changes, as well as shadows, specularities or texture changes. Clique no link abaixo para buscar o texto completo deste trabalho na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: