An electromyography database for determining optimal electrode placement for gesture recognition

We present a dataset with electromyography and inertial measurement data that contains common hand movements (relax, wave in, fist, wave out, and fingers spread), finger gestures (stretch index finger, victory sign, pistol sign, number four, and thumb up) and four functional grips (hold a mug, pick up coaster, hold paper cup, and hold a pen). The data set can be used for gesture recognition experiments.

Ethics:
The Medical Ethics Review Board of the University Medical Center Groningen issued a waiver (M22.307087) for the study in accordance with the Dutch Medical Research Involving Human Subjects Act. The study was performed in accordance with the standards for research involving human subjects defined by the Declaration of Helsinki. All participants gave written informed consent prior to data collection.

Participant Protocol:
All participants performed the gestures using their non-dominant hand. The experiment was divided into two parts. In the first part, the recording units—comprising inertial measurement units (IMUs) and surface electromyography (sEMG) sensors—were placed on eight forearm muscles ( (Supinator, Pronator Teres, Extensor Carpi Radialis Longus, Flexor Carpi Ulnaris, Extensor Carpi Ulnaris, Flexor Carpi Radialis, Extensor Digitorum Communis, Flexor Digitorum Superficalis)). These recordings are called anatomical mode as the sensors were placed directly over the corresponding muscles. In the second part, sEMG sensors were placed randomly around the circumference of the forearm (random mode). For each sensor placement mode participants performed the 14 distinct hand gestures ten times. Due to a technical issue, the order of the sensors in random mode were is unknown for 32 participants.

Acquisition Setup:
All data were recorded using the Trigno Discover system (version 1.6.4, Delsys Europe Ltd., United Kingdom). The sEMG acquisition system included four Delsys Trigno Avanti sensors and one Delsys Trigno Quattro sensor, along with a base station. All sensors utilized adaptive dry electrodes and operated wirelessly to allow unrestricted movement during recording. Each Delsys Trigno Avanti sensor had one sEMG electrode and one six-axis IMU. The sEMG signals recorded with the Avanti sensors were sampled at 4000 Hz with a voltage range of 5.5 mV. The embedded IMU recorded accelerometer and gyroscope data at 74 Hz. The Quattro sensor had four sEMG electrodes and a six-axis IMU. sEMG signals were sampled at 2000 Hz with a 5.5 mV range. Accelerometer and gyroscope data were collected at 148 Hz. For both sensor types, the accelerometer and gyroscope ranges were configured at ±2 g and ±250 degrees per second, respectively.
Video recordings used for labeling were captured with Azure Kinect (Microsoft Ltd., United States) and annotated using Adobe After Effects 2023 (version 23.6.0, Adobe Inc., United States) and Azure Kinect Viewer (version 1.4.1, Microsoft Ltd., United States). For privacy reasons the videos are not included in the public data set. The label files contain the observations of the scorers who focused on determining the beginning and the end of a gesture.