SMARTWEAR: AN INTELLIGENT OUTFIT RECOMMENDATION SYSTEM FOR WEATHER AND LIFESTYLE ADAPTATION

Abstract

This paper presents an outfit recommendation system that is weather and event context-aware and privacyoriented, aiming at making decisions more straightforward. To accomplish such a purpose, the use of a hybrid architecture based on computer vision techniques and lightweight machine learning is leveraged to allow real-time performance with a reduced computational load. Body measurement extraction is done through the application of pose landmarks and segmentation methods, while the age and gender of users are identified by means of quantized models running with TensorFlow Lite. Weather information and events data obtained from OpenWeatherMap and Google Calendar, respectively, are used to adapt suggestions to current and predicted environmental conditions. In particular, outfit suitability is calculated based on a weighted heuristic score, giving higher weights to event context (50%), weather information (30%), age (15%), and time (5%) considerations. The experiments conducted on 600 outfit scenarios have demonstrated the ability of the proposed model to generate suitable recommendations with a precision of 92.4% and a reduction in decision-making time of 73%.