Curriculum
Computer Science for Societal Challenges and Innovation, XXXVI series
Grant sponsor
Dipartimento di Matematica, UNIPD
Supervisor

Lamberto Ballan
Co-supervisor
s
Alberto Testolin

Project: Human Action Anticipation: Deep Learning Approaches Across Diverse Domains
Full text of the dissertation book can be downloaded from: not available yet

Abstract:Human action anticipation holds fundamental importance across various domains and applications. Anticipating human actions enables proactive decision-making, enhancing efficiency, safety, and overall performance of many systems, including robotic assistance systems, advanced surveillance systems, and autonomous driv- ing, where self-driving cars should be able to anticipate pedestrians’ intentions and actions to guarantee people’s safety. In this dissertation, our primary focus centers on anticipating human actions within two important domains, represent- ing the two main formulation of action anticipation: 1) Egocentric anticipation of in-kitchen activities; 2) Third-person anticipation of pedestrian actions. However, our research extends to cover the anticipation of the collective behavior patterns in traffic flows, and we extended even further to tackle the domain of abnormal behaviors decoding and recognition.