Estimating the Thumb Rotation Angle by Using a Tablet Device With a Posture Estimation Artificial Intelligence Model.

Area of Science:

• Computer Vision
• Biomedical Engineering
• Machine Learning

Background:

• Accurate hand pose estimation is crucial for various applications.
• Quantifying thumb rotation angle presents unique challenges in biomechanics.

Purpose of the Study:

• To evaluate the efficacy of combining MediaPipe with machine learning models for estimating thumb rotation angle.
• To compare the performance of different machine learning models in this task.

Main Methods:

• Utilized MediaPipe for hand landmark detection from video recordings of 15 healthy volunteers.
• Trained and evaluated four machine learning models: Linear Regression, ElasticNet, Support Vector Machine (SVM), and LightGBM.
• Measured thumb rotation angle (θ) against a palmar plane using an angle measuring device.

Main Results:

• The Support Vector Machine (SVM) and LightGBM models achieved the highest accuracy, with a correlation coefficient of 0.99.
• SVM and LightGBM models reported a root mean square error (RMSE) of 4.7 and 4.58, respectively.
• Linear Regression and ElasticNet models showed lower performance, with correlation coefficients of 0.91.

Conclusions:

• Combining MediaPipe with machine learning models enables highly accurate estimation of thumb rotation angle.
• The Support Vector Machine (SVM) and LightGBM models are particularly effective for this AI-driven biomechanical analysis.