Human Motion Tracking with Less Constraint of Initial Posture from a Single RGB-D Sensor

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Robotics

Background:

• High-quality 4D human motion reconstruction is crucial for immersive virtual reality (VR) and human operation.
• Existing methods face limitations due to self-scanning constraints and strict tracking requirements in monocular settings.

Purpose of the Study:

• To develop a human motion capture system that overcomes self-scanning constraints using only a single RGB-D sensor.
• To achieve high-fidelity 4D human motion reconstruction with improved ease of use and reliability.

Main Methods:

• Proposed a system combining human priors and performance capture with a single RGB-D sensor.
• Initialized complete mesh generation from front-view input to bypass self-scanning limitations.
• Updated the model during motion capture to maintain high fidelity and system usability.
• Integrated human priors to enhance the reliability of model warping.

Main Results:

• The system successfully generates a complete mesh from front-view input, addressing self-scanning constraints.
• The method allows for comfortable use while maintaining credible geometric warping.
• Demonstrated freedom from self-scanning constraints in extensive experiments.

Conclusions:

• The proposed system offers a more comfortable and effective solution for 4D human motion capture.
• It successfully overcomes the inherent self-scanning constraints of previous methods.
• The integration of human priors enhances the reliability of motion reconstruction.