Pedestrian Group Activity Recognition for Autonomous Vehicles and Robots: A Survey and Perspectives

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Human-Machine Interaction

Background:

• Pedestrians frequently appear in groups in human-machine interaction scenarios.
• Pedestrian groups offer richer data than individuals, aiding in resolving occlusion issues.
• The inherent randomness and spatiotemporal complexity of pedestrian actions make Pedestrian Group Activity Recognition (PGAR) difficult.

Purpose of the Study:

• To provide a comprehensive overview of the PGAR task specifically for autonomous vehicles and robots.
• To define pedestrian groups and activities within the context of autonomous systems.
• To systematically review existing datasets and methodologies in PGAR.

Main Methods:

• Systematic literature review of PGAR datasets and methods.
• Analysis of unique challenges and emerging trends in PGAR for autonomous driving and robotics.
• Definition of key terms: pedestrian group and pedestrian activity for autonomous systems.

Main Results:

• A novel definition of pedestrian groups and activities for autonomous systems is presented.
• A systematic summary of current PGAR datasets and methodologies is provided.
• Unique challenges and future trends in PGAR for autonomous vehicles and robots are outlined.

Conclusions:

• This article bridges a critical gap by focusing specifically on PGAR in autonomous driving and robotics.
• It serves as a foundational reference for researchers entering the PGAR field.
• The work highlights the importance of PGAR for safe and efficient human-machine collaboration.