Human Pose Estimation from Video and IMUs.

Area of Science:

• Computer Vision
• Robotics
• Human-Computer Interaction

Background:

• Video-based motion capture suffers from tracking artifacts due to image ambiguities, rapid motions, occlusions, and noise.
• Inertial sensors provide accurate limb orientation but lack continuous position information.
• Existing methods often struggle with drift or require extensive sensor setups.

Purpose of the Study:

• To develop a robust and stable full-body human motion capture system.
• To fuse complementary strengths of video and inertial sensor data.
• To overcome the limitations of individual sensor modalities for improved motion tracking.

Main Methods:

• A hybrid tracking approach combining video data with sparse inertial sensor orientation data.
• Utilizing video for drift-free and accurate position estimation.
• Employing inertial sensors for precise limb orientation and high-speed motion performance.

Main Results:

• Demonstrated improved performance and stability in human motion tracking.
• Successfully compensated for video tracking artifacts using inertial data.
• Achieved accurate full-body motion capture even during rapid movements.

Conclusions:

• The proposed hybrid approach significantly enhances human motion capture accuracy and stability.
• Fusion of video and inertial data offers a robust solution for challenging motion tracking scenarios.
• This method provides a more reliable alternative for applications requiring precise human motion analysis.