Towards efective and injury-free body building, we seek to explore the use of Electromyography (EMG) sensor for the detection of muscle compensation, a phenomenon often observed in physical exercises due to malcompliance or fatigue.
Weight training is one of the most popular modern exercises to build
muscle strength and resistance. It benefits not only the physical but
also mental health, where stronger muscles allow body movement
at a higher caliber, and being in shape raises also self-confidence
and social attention. Muscle building appears straightforward,
involving lifting, pulling, or pressing of weights, and more repetitions
or heavier weights result in more robust muscles. The dilemma is
however that incorrect postures or lifting excessive weights might
compromise the training efectiveness and, in the worst case, lead
to injuries [
To this end, we find opportunities in utilizing electromyography
(EMG) signals collected from widespread body surface to assess
muscle compensation. EMG signaling is a common and efective
method to evaluate muscle performance. Past works have focused
the use of EMG signals from a few targeted muscles and in the
laboratory settings [
Copyright 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
The surface EMG (sEMG) refers to the sensing of biopotentials
extracted from the electrodes on skin surface, and it has been used
widely for sports analysis, physical rehabilitation and as means
of human-computer interface. In particular, Ke Ma et al. [
Our hardware consists of a microcontroller, two muscle sensors, and a Bluetooth module. We use an Arduino Uno 1 as the microcontroller for the prototype. The EMG signal is extracted from Myoware 3-lead muscle sensors 2. We choose this particular muscle sensor because it provides not only raw EMG signals but also rectiifed and integrated EMG signals, which works well with Arduino’s analog-to-digital converter. The gain of signal rectification is adjustable on-board. We also connect an additional HC-06 Bluetooth module 3 for data transmission. 4
A small number of participants are invited to the prototype trial. As depicted in Figure 3, each participant is instructed to perform the bicep curl with dumbbells of specific weights. Within the participants’ capabilities, 5 weights, 2 kilograms apart are lifted. The 1https://store.arduino.cc/usa/arduino-uno-rev3 2http://www.advancertechnologies.com/p/myoware.html 3https://components101.com/wireless/hc-06-bluetooth-module-pinout-datasheet participants perform each set of exercise with constant speed within a 20 second duration. Figure 2 shows the rectified sEMG signals of one participant lifting 6kg and 8kg weights. The blue lines display the signals of the bicep muscles, on which five clusters that correspond to five sets of bicep curl are clearly visible. The red lines display the signals of upper trapezius muscles. As the participant lifts a 6kg weight, this signal remains at a relatively low and stable level. On the contrary, when lifting an 8kg weight, one can observe apparent peaks in the signal, which implies the substantial use of the trapezius muscles. This showcases the plausibility of an EMG-based wearable for muscle compensation assessment.
Furthermore, Figure 4 shows the normalized peak values of trapezius muscles of eight diferent participants. The peak EMG amplitude signals of the trapezius muscles in each exercise cluster are divided by the mean EMG amplitude signal during rest. As the weight increases, this normalized ratio increases as well. Generally, a surge of ratio occurs at 6kg and above, whereas individual diferences can be observed. (1) The yellow and sky blue lines, which increase slightly and stably, are collected from athletic participants. (2) The brown and grey lines, which show a surge at 4kg but remain flat through 6kg, are the only data collected from male participants. (3) The dark blue line shows a unique pattern– not increasing much through 8kg but shooting up at 10kg, which suggests that the participant might have no strength left at all for lifting the 10kg weight. These distinctions imply the distinctive physical abilities of individual participants, thus various extents of muscle compensation.
As for further utilization of the results, we find that there is no binary classification but rather multiple levels of muscle compensation. It is also more sensible to take into account the change of (i.e, the slope) of signal ratios as well as the sheer amplitudes in characterizing types of muscle compensation.
We are currently pursuing work in two dimensions. First is computerization of the process to quantify the level of muscle compensation and to characterize the risk of posture deformation, excessive weight, or fatigue. In the meantime, we seek to stream line the computation tasks on board the epidermal wearable to enable real-time user feedback. This may require an upgrade to Raspberry Pi or processors of higher computation capability.