Enhancing robustness of hand detection process for automated classification of tabletop activity videos

Abstract

Sharing information among prosthetic hand users and rehabilitation personnel is essential to improve the effectiveness, efficiency, and satisfaction of rehabilitation services that meet the user’s needs. The goal of this research is to establish an online platform for sharing video of pediatric prosthetic limb tasks and skills. The project's objective is to confirm a suitable video filming and processing factor and condition for automated task labeling of the video files. An orthogonal array experiment was conducted using four factors that affect hand detection. Comparison of the estimated and confirmed experimental gains showed that the Signal-to-Noise Ratio was better than the estimated value’s ± 30 %, hence, reproducibility was not confirmed with the control factor setting. Sensitivity was below ±30 % of the estimated value, and reproducibility was confirmed. Although the best condition in this experiment was approximate to the target output, further selection of control factors is required to suppress the variation.