Identificação de pessoas baseada em características antropométricas e de marcha extraídas de poses 2D
Carregando...
Data
2021-07-30
Autores
Tavares, Henrique Leal
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
De acordo com evidências atuais, em determinados cenários e ambientes, as características biométricas tradicionais, tais como impressões digitais, íris e faces, podem ter suas eficácias diminuídas na identificação de pessoas devido às condições heterogêneas ou precárias de iluminação dos ambientes, à distância das pessoas aos sensores de captura, ao uso de sensores de baixa resolução, à necessidade de cooperação das pessoas com o processo de identificação, dentre outros. Tendo em vista esta problemática, este trabalho propõe um método de identificação de pessoas baseado na fusão de características antropométricas e de marcha, classificadas como características biométricas soft e comportamentais, respectivamente, como uma forma de eludir-se destes obstáculos. O método proposto baseia-se em medidas de partes do esqueleto humano, como braços e pernas, e em mapas térmicos das articulações dos esqueletos, obtidos de poses 2D estimadas nas imagens por meio de algoritmos estado da arte propostos para esta finalidade. O método proposto foi avaliado na CASIA Gait Database-A e CASIA Gait DatabaseB, duas bases de dados de acesso público comumente utilizadas em pesquisas sobre identificação biométrica baseada em reconhecimento de padrões de marcha humana, e também em uma base de dados privada. Os resultados experimentais obtidos mostram que as características biométricas propostas, obtidas de poses 2D, podem ser utilizadas com êxito para a identificação humana, alcançando taxas de acurácia similares ou superiores às obtidas por métodos estado da arte.
According to current evidence, in certain scenarios and environments, traditional biometric characteristics, such as fingerprints, irises and faces, may have their effectiveness reduced in identifying people due to heterogeneous or poor lighting conditions of environments, distance from people to capture sensors, use of low-resolution sensors, need for people to cooperate with the identification process, among others. In view of this issue, this work proposes a method of identifying people based on biometric characteristics classified as soft biometrics, such as anthropometric measures, and behavioral measures, such as gait, as a way to overcome such obstacles. The proposed method is based on measurements of parts of the human skeleton, such as arms and legs, and on thermal maps of skeletal joints, obtained from 2D poses estimated in the images using state-of-the-art algorithms proposed for this purpose. The proposed method was evaluated in the CASIA Gait Database-A and CASIA Gait Database-B, two publicly available databases commonly used in research on biometric identification based on recognition of human gait, and also in a private dataset. The results obtained show that the proposed biometric characteristics, obtained from 2D poses, can be successfully used for human identification, reaching accuracy rates similar or superior to those obtained by state-of-the-art methods
According to current evidence, in certain scenarios and environments, traditional biometric characteristics, such as fingerprints, irises and faces, may have their effectiveness reduced in identifying people due to heterogeneous or poor lighting conditions of environments, distance from people to capture sensors, use of low-resolution sensors, need for people to cooperate with the identification process, among others. In view of this issue, this work proposes a method of identifying people based on biometric characteristics classified as soft biometrics, such as anthropometric measures, and behavioral measures, such as gait, as a way to overcome such obstacles. The proposed method is based on measurements of parts of the human skeleton, such as arms and legs, and on thermal maps of skeletal joints, obtained from 2D poses estimated in the images using state-of-the-art algorithms proposed for this purpose. The proposed method was evaluated in the CASIA Gait Database-A and CASIA Gait Database-B, two publicly available databases commonly used in research on biometric identification based on recognition of human gait, and also in a private dataset. The results obtained show that the proposed biometric characteristics, obtained from 2D poses, can be successfully used for human identification, reaching accuracy rates similar or superior to those obtained by state-of-the-art methods
Descrição
Palavras-chave
Identificação de pessoas, Biometria, Poses 2D, Mapas de calor das articulacoes, Medidas antropometricas, Reconhecimento de marcha, Peole identification, Biometric, Soft biometrics, Joints heat-maps, Anthropometric measurements, Gait recognitio