| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1.69 MB | Adobe PDF |
Advisor(s)
Abstract(s)
Este trabalho apresenta os resultados preliminares do desenvolvimento de um algoritmo de reconhecimento facial em tempo real através de redes neurais convolucionais (CNNs) .Elementos faciais, nomeadamente, rosto, boca, nariz e olhos, são detectados pelo algoritmo de Viola-Jones. Cada elemento facial é utilizado para treinar uma CNN. Os resultados de treinamento mostram uma acurácia de identificação de 100%. Testes em tempo real demonstram necessidade de aprimoramento. A base de imagens será futuramente ampliada para realização de um rigoroso procedimento de treinamento e teste do algoritmo.
This work presents the preliminary results of the development of a real-time facial recognition algorithm through convolutional neural networks (CNNs). Facial elements, namely, face, mouth, nose and eyes, are detected by the Viola-Jones algorithm. Each facial element is used to train a CNN. The training results show 100% identification accuracy. Real-time tests demonstrate a need for improvement. The image base will be expanded in the future to carry out a rigorous procedure for training and testing the algorithm.
This work presents the preliminary results of the development of a real-time facial recognition algorithm through convolutional neural networks (CNNs). Facial elements, namely, face, mouth, nose and eyes, are detected by the Viola-Jones algorithm. Each facial element is used to train a CNN. The training results show 100% identification accuracy. Real-time tests demonstrate a need for improvement. The image base will be expanded in the future to carry out a rigorous procedure for training and testing the algorithm.
Description
Keywords
Reconhecimento facial Tempo real Rede neural convolucional Algoritmo de Viola-Jones Face recognition Real time Convolutional neural network Viola-Jones algorithm