Full-Body Pose Tracking-The Top View Reprojection Approach.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Robotics

Background:

• Advancements in low-cost depth cameras have spurred progress in human body pose extraction.
• Current methods struggle with complex scenarios like multiple interacting users and severe occlusions.

Purpose of the Study:

• To present a model-based approach for markerless articulated full body pose extraction and tracking in RGB-D sequences.
• To address limitations of existing methods in handling occlusions and multiple users.

Main Methods:

• Employs a cylinder-based model for human body representation.
• Utilizes a Particle Filter to generate and track hypotheses for each body part.
• Introduces a novel metric using the reprojected Top View and depth information for hypothesis evaluation.

Main Results:

• The proposed method effectively handles severe occlusions and ambiguous cases.
• Experiments using public and custom datasets demonstrate robust performance.
• Comparative analysis against Microsoft's Kinect SDK and NiTE shows superior effectiveness.

Conclusions:

• The developed approach offers a significant improvement for markerless human body pose tracking.
• It provides a robust solution for complex, real-world scenarios with occlusions and multiple users.
• The novel evaluation metric is key to handling challenging tracking conditions.