Filtered pose graph for efficient kinect pose reconstruction.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Robotics

Background:

• Microsoft Kinect is widely used for marker-free motion capture in applications like industrial training and ergonomics.
• Kinect's accuracy is limited by placement requirements and susceptibility to occlusions.
• Real-time data-driven pose reconstruction aims to improve Kinect's robustness by using pose databases.

Purpose of the Study:

• To develop a more robust and accurate pose reconstruction method for marker-free motion capture systems like Kinect.
• To address the challenges of sub-optimal Kinect placement and occlusions in industrial environments.

Main Methods:

• A novel pose reconstruction method utilizing a Filtered Pose Graph to model intrinsic pose correspondences within a database.
• The Filtered Pose Graph optimizes the selection of relevant poses for efficient and accurate reconstruction.
• The method was tested in a challenging industrial setting with occlusions and non-ideal Kinect placement.

Main Results:

• The proposed Filtered Pose Graph method significantly speeds up pose database selection.
• Improved relevance of selected poses leads to higher quality pose reconstruction.
• Experimental results demonstrate superior accuracy compared to existing Kinect pose reconstruction methods in challenging conditions.

Conclusions:

• The Filtered Pose Graph is an effective approach for enhancing the accuracy and robustness of Kinect-based pose reconstruction.
• This method offers a significant improvement for real-time motion-based applications, particularly in industrial contexts with occlusions.
• The approach provides a more reliable solution for user training and ergonomics evaluation using Kinect technology.