Personalized In-Store Advertising using Deep Learning for Face Analysis and Gaze Estimation

Advertising is an essential part of our current economy and there is a clear demand for solutions that target users specifically based on their needs and preferences. Users prefer the tailored content, and advertisers reduce costs by reaching relevant audiences from their target customer segment. This problem is mainly addressed in online environments, but less so in the physical world. In this paper, we present a deep learning-based system for delivering targeted advertisements to customers in supermarkets or large stores, using tablets dispersed through the aisles. The system leverages multiple neural network models for real-time face detection, recognition, and emotion detection to curate and display relevant ads. By incorporating gaze estimation and emotion tracking, the system assesses user engagement with the advertisements, enabling continuous improvement in ad targeting and personalization. We perform a comprehensive quantitative analysis to evaluate various deep learning models for face recognition and gaze estimation in diverse, realistic scenarios. Our proposed system employs the models that demonstrate superior accuracy and speed in our experiments, ensuring optimal performance in dynamic environments. Our approach offers a significant advancement in physical world personalized advertising, blending state-of-the-art deep learning techniques with practical application in real-world settings.