Desenvolvimento de placa sensora instrumentada com redes de Bragg em fibras óticas para autenticação biométrica

Abstract

This work presents the design, development, and results of a sensing board instrumented with fiber Bragg gratings, detailing the steps of fabrication and testing for the purpose of biometric authentication from the footstep patterns of users. The fiber Bragg gratings were encapsulated in acetic curing silicone and positioned in regions of the board to build sensing areas corresponding to the size of adult feet to allow data acquisition in a static regime. From the measured data, a system response model was developed, as well as a graphical representation through normalized pressure maps. Two pattern recognition models were also developed based on machine learning and deep learning techniques, generating a support vector machine model and a convolutional neural network capable of classifying different footstep patterns for biometric authentication purposes. The trained support vector machine achieved accuracy rates above 90% and 86% for unilateral and bilateral training, respectively. The neural model, always trained with 10 epochs, was able to get more than 77% of the samples correct for unilateral training, while bilateral training showed accuracy greater than 96%.