The Research Of Human-Computer Interaction Based On Facial EMG Signal

In recent years,by decoding human bioelectrical signals to give machines the ability to understand human intentions,it has become a hot research direction in the field of human-computer interaction.Because of its high signal-to-noise ratio and easy acquisition,EMG signals have received wide attention and have been applied in human-computer interaction based on gestures,mouth patterns and other recognition.Hhis paper studies human-computer interaction technology based on facial EMG signals.Because different facial movements(such as frown,blink,bite etc.)require different levels of muscle group participation,resulting in differences in position amplitude,frequency,etc.of facial EMG signals.Choosing the proper facial movement,using signal processing and machine learning techniques to accurately detect and identify the difference in EMG pattern,can accurately recognize different facial movement,and thus output the recognition result as an external interaction command,realizing a new human-computer interaction system without hands and voice.This paper focuses on the accurate real-time recognition algorithm for a variety of facial movements with common EMG patterns.The main works are:1)Research on facial EMG signal pattern recognition algorithm based on forehead single channel.Through experimental analysis,Six kinds of facial movements were chosen,which have common EMG signal patterns through different people and different time.We design a recognition algorithm with short-time energy data segmentation method,multiple time-domain features and random forest classifier.The results show that this method can achieve an average correct rate of 92.05%in 10 subjects.This method with cross-subjects transfer learning ways can further improve recognition accuracy,reaching 94.17%with data augmentation and achieving 84.25%recognition accuracy without self-training data.2)Research on multi-channels facial EMG signal pattern recognition algorithm.A real-time feature extraction method is proposed,which combines time domain features in channels,time domain features and correlation features between channels,with random forest classification algorithm to identify 16 facial movement,achieving an average recognition accuracy rate of 96.8%.And we propose a facial movement combination method can theoretically extend the number of interactive command to 60,but requires a large training cost.In this paper,three users tested 30 kinds of combinations of facial movement.Through training,it achieves an average accuracy of 83.9%.In addition,in order to decrease recognition error caused by head movement interference,the gyro sensor's data is used to track the head movement posture changes and feedback to the recognition algorithm,which increases the robustness of the interactive system.3)Develop some real-time human-computer interaction test systems based on single-channel and multi-channels facial EMG signal pattern recognition method,including driving assistance system,robotic arm control system,UAV control system,and typing system.And the test results proved the effectiveness and practicability of these systems.