Método de reconhecimento da marcha humana por meio da fusão das características do movimento global

Milene ArantesAdilson Gonzaga

This paper proposes a novel computer visionapproach applied to video sequences of people walking torecognize them by their gait. Human motion carries differentinformation considering different ways of analysis. Theskeleton carries motion information about human joints andthe silhouette carries information about boundary motion ofthe human body. The binary and gray-level-images containdifferent information about the human motion. Our proposalis to recover these different kinds of information of the humanbody based on four different segmented image models, andapllies a fusion model to improve classification. The basicidea is to consider frames as elements of distinct classes ofvideo sequences and the sequences themselves as classes ina database. The methodology applies background extractionusing the Gaussian Mixture Model (GMM), a scale reductionusing a multiresolution technique based on the Wavelet Transform(WT) and feature extraction using Principal ComponentAnalysis (PCA). The Classification rates obtained separatelyfrom four different kinds of image sequences are them mergedtogether by a fusion technique.

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: