Enhancing the Accuracy of A-ROM Measurements for Finger Joint Angles using Image Processing and Automated Computation
Huu-Hieu Quang, Yoshifumi Morita, Noritaka Sato, Makoto Takekawa · Engineering Technology & Applied Science Research · 2025
In our previous research, we developed a technique for quantifying finger joint angles to assist rehabilitation therapists in assessing patients' Active Range of Motion (A-ROM). Therapists often face the challenge of accurately adjusting camera orientations (roll, yaw, pitch) to precisely capture finger joints. Our previous approach, a deep learning-based "posture determination method," required extensive training data to achieve accuracy. Moreover, accurately determining the 3D centerline of a finger proved difficult due to the limitations of the Hand Keypoint (HKP) method. Deviations in the camera pitch angle from 0° introduced distortions in the Point Cloud Data (PCD), necessitating repeated captures and thus reducing efficiency without effectively addressing the distortion. To address these limitations, we introduce several methodological advancements in this study. First, we implement a new "posture determination method" that utilizes image processing to reduce the reliance on large training datasets. Second, we enhance the accuracy of the centerline measurement by developing a calculation method that utilizes joint positions derived from image processing. Lastly, we establish a technique to correct distortions in the PCD of the centerline measurement, including adjusting the pitch angle back to 0°, thereby compensating for any data distortion. The new method was validated on five healthy participants and achieved a mean absolute error of 1.3° in the measurement of finger joint angles, which satisfies the accuracy requirement of 2°. In addition, the average measurement time was significantly reduced from 4.9 s with the previous method to 1.3 s.