Towards Eating Habits Discovery in Egocentric Photo-Streams

Eating habits are learned throughout the early stages of our lives. However, it is not easy to be aware how food-related routine affects healthy living. In this work, we address unsupervised discovery nutritional from egocentric photo-streams. We build a behavioral pattern model, which discloses routines activities performed days. To do so, rely on Dynamic-Time-Warping for evaluation similarity among collected Within framework, present simple, but robust and fast novel classification pipeline that outperforms state-of-the-art image with weighted accuracy F-score 70% 63%, respectively. Later, identify days composed describe person as anomalies in daily life user Isolation Forest method. Furthermore, show an application identification scenes when camera wearer eats isolation. Results have shown good performance proposed model its relevance visualize individuals.