Facial Image-based Age and Gender Classification to Enhance Recommendations in a Smart TV Environment

Advances in Machine Learning and Deep Learning have made it possible to do a myriad of tasks that were either too complex to hardcode or even deemed impossible. We have taken leaps in security, medicine, linguistics and accessibility with their help. Day by day, the number of appliances using these mechanics is increasing, with new vehicles to smart homes implementing these changes. This revolutionary technology can be utilized to detect and predict one’s gender and approximate age through the lens of a camera either in the form of a photo or through live video. This information is vital in privacy and security aspects. However, it can also be applied to the overlooked emergent device, the smart television. We want to stress the importance of the data of age and gender in today’s television climate and how it can be implemented to benefit our day-to-day life hence why we decided to use such technology to aid in a common but often overlooked task. There are on average 850 channels that can be browsed in our very own television and therefore it can be a hassle to find the channels we would like to watch. We wish to tackle this problem by using an estimated combination of deep learning and a classification-based machine learning model which will be integrated with an IoT device (microcontroller) all of which is linked by a data pipeline. By training our multi-component model on a dataset on facial images along with a custom dataset regarding the preferences of channels in different age and gender groups, we can help solve such a task.