Chi-Jui Wu, Shen-Jhe Haung and Chien-Chang Chen
The purpose of rehabilitation is to recover the degradation of human body strengths and flexibility caused from damaging by disease or injury. Improper posture practices in a long-term rehabilitation, however, might cause secondary injuries. To avoid these predicaments, the paper presents a machine learning algorithm for motional pattern recognition of various upper limb motions and wrist rotations. The proposed hardware scheme included single triple-axis accelerometer and a personal smart device. The accelerometer was setup on subject’s wrists to detect and then record the time-dependent sequential signals, and the acquired datasets were downlinked to the personal smart devices. The information of time sequence of limb motions was then analyzed using our proposed algorithm. The main data cluster numbers were estimated using data density functional method, and locations of data centroids were then measured using the Gaussian mixture model. Thus, swing angles of the limb motions can be further analyzed using the combinational machine learning algorithm. Under the proposed experimental framework, swing angles of the limb motion and wrist rotation can be clearly measured even though the motions of subjects were unstable. Then the feedback results fed to the time sequences can assist the posture corrections. Therefore, the technique can be used for analyses of accident circumstances and then dangerous alarms. In a nutshell, the proposed framework not only provides highly plausibility and objectivity but also reinforces the commercialization.
PDFShare this article
