1) Electrodes (surface or needle) record the potential differences generated by recruited motor units. 2) The weak signal is amplified and sampled (analog-to-digital conversion). 3) The signal is filtered and cleaned of artifacts (crosstalk, motion, mains). 4) Features are extracted (e.g. RMS, MAV, spectral features) and then classified (e.g. by a machine-learning model) to recognize a gesture or derive a control command.
Enables reading motor intentions directly from the muscles, even before a full movement occurs, providing a natural, touch-free control signal for prosthetics, exoskeletons and robots.
Sensors recording muscle potentials: surface (sEMG) on the skin or needle inside the muscle.
Official
Amplifies and digitizes the weak myoelectric signal.
Official
Filtering, artifact removal and feature extraction (RMS, MAV, spectrum) from the EMG signal.
Official
Signal from neighboring muscles (crosstalk) and electrode shift reduce gesture-recognition accuracy.
EMG patterns depend on the person, posture and fatigue, requiring calibration and adaptation.
Étienne-Jules Marey makes early recordings of muscle electrical activity and introduces the term electromyography.
Adrian and Bronk develop motor-unit recording and the concentric electrode, enabling clinical EMG.
Surface EMG becomes popular in wearable armbands (e.g. Myo) for touch-free control and human-machine interfaces.
Time complexity: O(kanaly × probki) na okno. Space complexity: O(kanaly × bufor).
Muscle crosstalk, electrode shift, sweating and inter-individual variability make reliable gesture classification difficult.
Surface (sEMG) vs needle.
Number of electrodes / measurement sites.
sEMG typically up to ~1-2 kHz (band ~20-450 Hz).
Continuous, parallel sampling of all channels.
Continuous signal acquisition, no routing.
Channels and features can be processed independently/in parallel.
EMG acquisition and processing are lightweight; they run on CPU/embedded devices. The key is dedicated analog hardware (electrodes + amplifier).
