Employing Shadows for Multi-Person Tracking Based on a Single RGB-D Camera

Area of Science:

• Computer Vision
• Robotics
• Human-Computer Interaction

Background:

• Multi-person tracking is challenging due to occlusions.
• Existing methods often fail when individuals are hidden from view.
• Shadows are typically discarded as noise in tracking algorithms.

Purpose of the Study:

• To develop a novel method for multi-person tracking that overcomes occlusion challenges.
• To utilize people's shadows as a viable tracking cue.
• To fuse shadow and skeleton data for robust tracking.

Main Methods:

• A novel method fusing shadow data from RGB images with skeleton data from an RGB-D camera.
• Utilizing shadow data to track individuals during full occlusions.
• Employing skeleton data for direct position capture.

Main Results:

• The proposed method successfully handles full occlusions in multi-person tracking.
• Experimental results confirm the efficiency of the shadow-fusion technique.
• Demonstrated substantial value in resolving occlusion problems.

Conclusions:

• Shadow data can be effectively used as a clue for multi-person tracking.
• Fusing shadow and skeleton data offers a robust solution to occlusion.
• The method shows potential for various camera types and tracking scenarios.