Object Class Detection in Omnidirectional Images

Fábio R. AmaralAnna H. R. Costa

In this paper we investigate the use of a generative modelto the object recognition task in omnidirectional images.The purpose of our work is to find and classify objects typicallyfound in indoor office environments (tables, chairs,etc) through the analysis of images obtained from an omnidirectionalvision system. First, a set of generic featuresare obtained from the query image and clustered in appearanceclusters. In training mode we make use of labeled featuresto compute the joint probability for the containingclasses by matching those features to the appearance clusters.The recognition proceeds by matching the features extractedfrom the query image to the model, computing thelikelihood used in the decision equation through the Bayesrule. Results are presented for some first experiments.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: