Identification and Verification Datasets Based on Surface Electromyography

During the experiment, the subjects stood still with their arms hanging down naturally. The subjects were asked to perform 3 gestures using only their right hand, with the forearm and upper arm remaining as still as possible. The 3 gestures as presented in Fig. 3 are the following: (1) static gesture: wrist flexion with the hand closed, (2) static gesture: index finger, middle finger, and little finger extension, and (3) dynamic gesture: wrist counterclockwise rotation with the hand closed - hand open - hand close (see Supplementary Video 1 demonstrating the 3 designed gestures). Before each trial, subjects had enough time to learn how to perform each gesture. During each trial, the instruction video was played to guide the subjects to perform gestures in accordance with the sequence. Each trial was compromised of a 10-second pre-trial resting period for getting ready, followed by 20 execution-rest pairs, and ended with a 5-second post-trial resting period., as shown in Fig. 4. Each execution-rest pair consisted of a 3-second gesture performance and a 3-second rest. Three trials of three gestures were performed separately. Between each trial, enough time was offered to rest. Wrong execution and malformed signals were removed from the database, and the collection was repeated until the database contained 20 complete sets of signals for each gesture of each subject. Two sessions with the same experimental procedure were acquired on different days. Data from 17 subjects were included in Session 1, and data from subjects 1 to 8 were included in Session 2 on another day (interval 2-77 days, 47.25±37.50 days on average). In Session 2, electrodes were replaced, thus the effect of minor electrode shift and minor gesture variation on personal identification and verification performance were considered in this study.