Denoising of Joint Tracking Data by Kinect Sensors Using Clustered Gaussian Process Regression.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Computer Interaction

Background:

• Kinect sensors offer potential for remote patient monitoring during rehabilitation exercises.
• Kinect's joint position data can be unreliable due to occlusions.
• Motion capture (MOCAP) systems provide accurate tracking but are expensive and inconvenient.

Purpose of the Study:

• To improve the accuracy of Kinect sensor joint position measurements for rehabilitation.
• To develop a cost-effective and convenient solution for reliable patient exercise monitoring.

Main Methods:

• Simultaneously captured Kinect and MOCAP data during a training phase.
• Trained a Gaussian Process regression model to map noisy Kinect data to accurate MOCAP data.
• Proposed a joint standardization method to normalize for variations in limb length and body posture.

Main Results:

• The proposed method significantly improved the accuracy of Kinect-derived joint positions.
• Denoised Kinect measurements were more accurate than benchmark methods.
• The joint standardization method effectively handled inter-person variability.

Conclusions:

• The developed method enhances the reliability of Kinect sensors for remote rehabilitation.
• This approach offers a more accessible and accurate alternative to traditional MOCAP systems.
• The findings support the use of improved sensor fusion techniques in healthcare technology.