Group behavior analysis in videos

• Main Author: Hu, Bo
• Other Authors: Cham Tat Jen
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2025
• Subjects: Computer and Information Science; Human behavior analysis; Trajectory prediction; Action prediction
• Online Access: https://hdl.handle.net/10356/182244
• Institution: Nanyang Technological University

In this thesis group behavior in videos is analyzed. The behavior of a person in a group is determined by both self-intention and social interaction. To investigate the self-intention and social interaction, three settings of group behavior are explored, which are behavior affected by only self-intention, only social interaction, and both self-intention and social interaction. Different tasks are selected for each setting, where several novel frameworks are proposed. The progress regression is designed to model the temporal relations in an action; entry-flipping mechanism is introduced to model the human-human interactions between target and observed participants; trajectory-scene-cell representation is proposed to convert trajectories and scene into one feature space and unify the human-human and human-scene interactions modeling. Overall, this thesis provided a detailed analysis on effects of self-intention and social interaction in group behaviors and proposed several approaches for multiple types of relation modeling in different tasks of group behavior analysis.