DisCaaS: Micro Behavior Analysis on Discussion by Camera as a Sensor.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Behavioral Analysis

Background:

• Commercial cameras aid human activity understanding.
• High angle-of-view cameras enable novel analysis perspectives.
• Micro-behaviors in meetings are often overlooked.

Purpose of the Study:

• To develop a system for quantified meeting analysis using a single 360-degree camera.
• To recognize micro-behaviors, specifically speaking and nodding, during group discussions.
• To evaluate the system's performance on diverse datasets, including virtual meeting scenarios.

Main Methods:

• Utilized a single 360-degree camera for data capture.
• Proposed a method to recognize speaking and nodding from video streams of face images.
• Employed a random forest classifier for micro-behavior recognition.
• Created three datasets, including physical and virtual meeting recordings.

Main Results:

• Achieved a macro average F1-score of 67.9% for speaking and nodding detection in cross-validation.
• Demonstrated a macro average F1-score of 62.5% in a leave-one-participant-out cross-validation.
• Evaluated the system's applicability to virtual video conferences.

Conclusions:

• The proposed system effectively recognizes speaking and nodding micro-behaviors in group discussions.
• The approach shows promise for automated and quantified meeting analysis.
• Further research is needed to address challenges in applying the system to virtual settings.