Human behavior understanding from motion and bodily cues using deep neural networks

Technological advancements in the field of Artificial Intelligence (AI) have opened the path to systems capable of learning and sensing the environment in a way that imitates human perception. Machines are very powerful when it comes to learning regular and tangible patterns. However, there is still big room for improvement in the fields concerning the automatic understanding of behaviors and how humans use them to communicate as well as to express their feelings. This dissertation poses the critical research question of how to build computational models that can enhance machines' understanding of human intentions, behaviors, personality traits, and activities, by learning meaningful patterns from human motion and bodily cues. The findings show that by examining the rich information conveyed by human nonverbal communication (e.g. gestures, body postures, and movements), we can build smart applications in critical fields of our society such as Healthcare, Surveillance, and Affective Computing.