Evaluating Automatic Body Orientation Detection for Indoor Location from Skeleton Tracking Data to Detect Socially...

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Social Robotics

Background:

• Analyzing social dynamics in indoor spaces requires understanding spatial-temporal variables and invisible social constraints.
• Current sensor systems primarily detect physically occupied spaces, neglecting the 'socially occupied' areas defined by interactions.
• F-Formation analysis is crucial for understanding social structures in small gatherings.

Purpose of the Study:

• To develop a system for detecting socially occupied spaces, not just physically occupied ones.
• To calculate body orientation and location using skeleton joint data from depth cameras.
• To anonymously and automatically assess social interaction dynamics in indoor settings.

Main Methods:

• Integrating depth cameras to capture skeleton joint data (shoulders, spine, head/face).
• Deriving body orientation by combining joint data with spatial-temporal trajectory information.
• Utilizing physically occupied measurements to infer socially occupied spaces.
• Conducting a user study comparing Kinect v2, Azure Kinect, and Zed 2i depth sensors.

Main Results:

• The system achieved high accuracy in assessing socially relevant body orientation: 90% (Kinect v2), 96% (Azure Kinect), and 89% (Zed 2i).
• The algorithm enables anonymous and automated assessment of socially occupied spaces.
• Depth sensor systems show promise for detecting complex social structures.

Conclusions:

• The proposed method effectively detects socially occupied spaces by analyzing body orientation and location.
• The system offers a valuable tool for research in group interactions within complex indoor environments.
• This technology has significant implications for understanding and analyzing human social behavior in real-world settings.