Dynamic Gesture Recognition Based on Arm Muscle Change Detection and Machine Learning Using Fiber Optic Sensor

Fiber optic sensor is capable of capturing tiny displacements and deformations, making them ideal for real-time monitoring of human limb movement, so it has great potential in human-computer interaction based on gesture recognition. In this paper, we propose a wearable, flexible polymer-embedded fiber Bragg grating sensor array to detect subtle changes in the brachialis and palmaris longus muscles during muscle contractions in order to recognize different dynamic gestures. Nine kinds of common dynamic gestures are considered. Four fiber optic sensors are firmly attached to the surface of the target muscle group with adhesive tape to detect tiny changes in the muscle. Changes in different muscle contractions cause some differences in the optical signals output by these fiber optic sensors. These optical signals are collected and segmented into signal segments corresponding to each dynamic gesture. We propose a method to convert multi-dimensional optical signal segments into 2D image. These images are fed to the Convolutional Neural Network (CNN) to classify different dynamic gestures. Traditional machine learning methods based on image feature extraction are compared. Experimental results show that CNN has a stronger ability to extract subtle differences in muscle changes from 2D image, which helps to classify different dynamic gestures, with an average recognition rate of 98%. This paper provides a new and valuable reference for hand motion monitoring.