Noncontact Measurement Technique for Simulated Hemiplegia Walking Motion using Ultra-high Sensitive Electrostatic Induction Sensor

In this study, simulated hemiplegia gait is detected by using a method that detects the subject's walking motion without attaching a sensor to the subject. To detect walking motion without contacting or wearing the human body, attention has been focused on the fluctuations in human body potential associated with walking motion. Walking motion is detected under noncontact conditions by installing an ultra-sensitive electrostatic induction current sensor near a walking subject and detecting changes in human body potential caused by the subject's walking motion. In this study, to simulate gait disorders, ankle weights of different weights are attached to the right ankle of a healthy subject, and the electrostatic induction current waveforms induced by the walking movements of four tasks are measured without contact. A scalogram is obtained by wavelet transforming the electrostatic induction current waveform induced by this walking motion. By using these scalograms as training data for deep learning, a learning model was constructed that can detect the inconvenience of a subject's walking motion. Convolutional neural network (CNN) and long short-term memory (LSTM) are used for deep learning. The results showed that when CNN and LSTM are used for deep learning classification, the average classification accuracy rate of both methods is approximately 80%. The results of this study reveal that when hemiplegia causes asymmetry in walking movement, it is possible to easily detect the degree of lameness of hemiplegia without attaching a sensor to the subject.