Título: Reconhecimento de Posturas Manuais Usando Redes Neurais
Autores: Lamar, Marcus V.; Bhuiyan, Md. Shoaib; Iwata, Akira
Resumo: In this work we present a fast and efficient feature extraction method to be applied in visual gesture recognition systems. The method is based upon the using of Principal Component Analysis (PCA) to extract morphological information about 2D regions. The system performs the modeling of hand postures from color gloved video images. The performance of the feature extraction method is evaluated in applications of Japanese and American finger spelling automatic recognition system. The use of color gloves allows a fast tracking and complex hand model be extracted against natural backgrounds. A feedforward multiplayer perceptron neural network classifier achieved 89.4% of correct recognition rate in a set of 42 Japanese fingerspelling postures, and 94.5% in a 26 hand postures set extracted from American fingerspelling.
Código DOI: 10.21528/CBRN2003-066
Artigo em PDF: 6CBRN_066.PDF
Arquivo BibTex: 6CBRN_066.bib