Semiring-Optic-Fiber (SROF) Sensor-Based Abnormal Gait Recognition via Monitoring Muscle Activation

With the rapid development of wearable robotics, the requirements for wearable sensors to detect the interaction between wearable robots and humans are increasing. This study proposed a noncontact bendable-sensitive sensor using a semiring optical fiber for monitoring muscle activity. Raw data were from seven subjects with five gaits (four abnormal gaits and one normal gait), and traditional machine learning, e.g., support vector machine (SVM) and
${k}$
nearest neighbors (
${k}$
-NNs), and big data-driven neural networks, e.g., convolutional neural networks (CNNs), recurrent neural networks (RNNs), and temporal convolutional networks (TCNs), were used to recognize five gaits, due to the complexity of gait-muscle models. Using SVM,
${k}$
-NN, CNN, RNN, and TCN, the average of best recognition accuracies of the proposed sensor was 86.8%, 92%, 96.9%, 99.9%, and 99.4%, respectively. The recognition results suggested that a semiring-optic-fiber (SROF) sensor contained potential information of muscle activity during five gaits, and compared with SVM and
${k}$
-NN, neural networks can better filter and extract features from raw data, and even RNN and TCN models can reach 100% accuracy for certain subjects. This work paves a new way for recognizing abnormal gaits using an SROF sensor.