Uma abordagem para analisar a atenção de espectadores utilizando visão computacional.

Abstract

DigitalSignageconsistsincontenttransmissionbydigitalmedia,veryoftenusinginformative panels displayed in public places for audience attention arouse. The research described in this dissertation, consists in creating a method for video display, environment recording, and attention modeling in video display for digital signage scenarios. In this study, computer vision tools were used to categorize audience by gender and age group. Bayesian Networks were automatically builtusing there cords obtained from transmittedads, in which the correct audience profile representation had area under the curve (AUC) of 0.82. When performing this research, it were observed gaps in validation of age classifiers using faces. To address these deficiencies, an age stratification approach was proposed to train a neural network that achieved superior performance in comparison with the state of the art which was 72% when validated on the MORPH face dataset. To make it possible to carry out this research, due to the lack of a publicy available dataset of labeled videos, a video dataset was created containing 152 videos manually labeled in six categories. The analysis of the faces of the viewer and the classification of such images with respect to gender and age indicated that the individuals belong to different categories that dedicated varying degrees of attention todifferentvideos. Thus,thesystemdevelopedcanbeusedinthecategorizationofthespectators, for automatically transmitted advertisements as a way to aid in the allocation of time for the transmission of advertising companies, among other applications.