Description

Gaze communication is a primitive form of human communication that plays an important role in augmenting verbal communication during social interactions.
Perceiving and identifying gaze communication in videos is crucial for social activities and social scene understanding.
Advances in automatic non-verbal communication understanding fro videos 1) would provide evidences for robot systems to learn human behavior patterns in gaze communication
and further facilitates interactions between humans and robots; 2) would enable simulation of more natural human gaze communication behaviors in Virtual Reality environment; 3) would help to evaluate and diagnose children with autism.

The project will consist of the following phases:
- extracting nonverbal communication cues from a dataset of videos of social interactions
- developing an approach based on Machine Learning and Geometric Deep Learning (Graph convolutional Networks) to understand gaze communication from videos.
- model evaluation and validation
- quantitative comparison to state of the art approaches