Towards Better Recognition of Age, Gender, and Number of Viewers in a Smart TV Environment
Abdul Haq, Shah Khusro, Iftikhar Alam · 2021
The retrieval of age and gender-related information from digital facial images has recently gained attention. This information is used for numerous purposes, including better Human-Computer Interaction (HCI), security, privacy, and recommendations on smartphones and computers. However, little attention has been given to utilize this information for emerging computing devices, i.e., smart TV. The detection of gender, age, and the number of viewers in smart TV watching environments is challenging due to several factors, such as watching distance and room lighting conditions. In this research, efforts have been made to develop a novel approach to accurately recognize numbers, genders, and age groups of Smart TV viewers based on human faces. A custom dataset of facial images representing the Smart TV room environment is developed and used to train our proposed model. Our proposed architecture model is based on Convolutional Neural Network (CNN). We found 91.1% accuracy of training for the gender classification and 79.1% for predicting age, specifically in a smart TV watching environment where watching distance and room lighting condition matters.