Predicting the future from first person (egocentric) vision: A survey

Egocentric videos can bring a lot of information about how humans perceive the world and interact with environment, which be beneficial for analysis human behaviour. The research in egocentric video is developing rapidly thanks to increasing availability wearable devices opportunities offered by new large-scale datasets. As computer vision techniques continue develop at an pace, tasks related prediction future are starting evolve from need understanding present. Predicting activities, trajectories interactions objects crucial applications such as human–robot interaction, assistive technologies both industrial daily living scenarios, entertainment virtual or augmented reality. This survey summarizes evolution studies context making overview applications, devices, existing problems, commonly used datasets, models input modalities. Our highlights that methods have significant impact range further efforts should devoted standardization proposal datasets considering real-world scenarios ones vocation.