Gesture Recognition Through the Implementation of a Bimodal

Acquisition System Using EMG and FMG Signals

Nuno Pires

and Milton P. Macedo

1,2 a

Instituto Politécnico de Coimbra, Rua da Misericórdia,

Lagar dos Cortiços, S. Martinho do Bispo, 3045-093 Coimbra, Portugal

LIBPhys, Department of Physics, University of Coimbra, Rua Larga, 3004-516 Coimbra, Portugal

Keywords: Bionic Hand, Electromyography, Force Myography, Feature Extraction, Gesture Recognition, Classification

Models.

Abstract: This study is part of a broader project, the Open Source Bionic Hand, which aims to develop and control, in

real time, a low-cost 3D-printed bionic hand prototype using signals from the muscles of the forearm. In this

work, it is intended to implement a bimodal signal acquisition system, which uses EMG signals and Force

Myography (FMG), in order to optimize the recognition of gesture intention and, consequently, the control of

the bionic hand. The implementation of this bimodal EMG/FMG system will be described. It uses two

different signals from BITalino EMG modules and Flexiforce™ sensors from Tekscan™. The dataset was

built from thirty-six features extracted from each acquisition using two of each EMG and FMG sensors in

extensor and flexor muscle groups simultaneously. The extraction of features is also depicted as well as the

subsequent use of these features to train and compare Machine Learning models in gesture recognition,

through MATLAB's Classification Learner tool. Preliminary results obtained from a dataset of three healthy

volunteers, show the effectiveness of this bimodal EMG/FMG system in the improvement of the efficacy on

gesture recognition as it is shown for example for the Quadratic SVM classifier that raises from 75,00% with

EMG signals to 87,96% using both signals.

1 INTRODUCTION

Upper limb myoelectric prostheses, also called bionic

hands, are electromechanical devices that are attached

to the residual limb of amputees, in order to replicate

the functionality of the human hand, and

consequently improve the quality of life of these

people.

Commercial bionic hand models use surface

electromyographic (EMG) sensors to capture the

electrical activity produced when muscle remnants

are activated. However, this is a detection method

whose effectiveness is susceptible to external

electromagnetic noise, muscle fatigue, or impedance

changes in the sensor-skin interface. So research in

the field of myoelectric prostheses is faced with the

constant challenge of replicating the functionality of

the human hand.

This study is part of a broader project, the Open

Source Bionic Hand, which aims to develop and

https://orcid.org/0000-0003-0595-5298

control, in real time, a low-cost 3D-printed bionic

hand prototype using signals from the muscles of the

forearm. In literature it is possible to find previous

contributions from this project, focused on the

implementation of a prototype of a low-cost

controller of a bionic hand, namely from the

application of alternative mechanomyographic

sensors and novel and low-cost electrodes, built from

a conductive leather material as well as based on

desktop 3D printing using conductive PLA (Pol-

yLactic Acid) (Marques, 2020) (Silva, 2019).

The main objective of the work presented in this

paper it is the implementation and evaluation of the

effectiveness of a bimodal EMG/FMG signal

acquisition system for the control of a bionic hand.

The idea is to counter the limitations of EMG sensors

by integrating FMG, which shows benefits such as

robustness in the face of impedance changes at the

1018

Pires, N. and Macedo, M. P.

Gesture Recognition Through the Implementation of a Bimodal Acquisition System Using EMG and FMG Signals.

DOI: 10.5220/0013401800003911

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2025) - Volume 2: HEALTHINF, pages 1018-1026

ISBN: 978-989-758-731-3; ISSN: 2184-4305

skin interface and sweating, and lower sensitivity to

sensor positioning. This is despite having its own

challenges, such as sensitivity to unintentional move-

ments and external noises.

The term FMG, or force myography, describes the

various non-invasive techniques that use force

sensors to detect voluntary changes associated with

the activation/deactivation of superficial muscle

groups relative to a default state that usually

corresponds to the limb in a relaxed position

(Grushko, 2020). It also detects voluntary changes

caused by the movement of tendons under the surface

of the skin (e.g. in the wrist) (McIntosh, 2016).

The first work on the FMG technique as a

modality for the control of myoelectric prostheses

was published in 1999 (Abboudi, 1999) but it was

only in the middle of the last decade that it gained

traction among researchers, driven by the

development of Machine Learning techniques.

Several scientific publications present promising

results on the possibility of using the FMG technique

to predict movement intention in implementations of

bionic hand prostheses (Citi, 2016) (Kadkhodayan,

2016) (Radmand, 2016). More recently, there is a

growing interest in combining sEMG and FMG in

order to create more robust control systems to be used

by pattern recognition models (Jaquier, 2017)

(Nowak, 2017) (Xiao,2017). What makes the bimodal

system interesting is the fact that it detects both

electrical and volumetric phenomena associated with

muscle contraction. In 2020, Jiang et al., proposed a

co-localized approach to acquire EMG and FMG sim-

ultaneously at the same location, achieving a 10%

increase in accuracy in identifying 10 American sign

language signals, relative to isolated modalities

(Jiang, 2020).

In general, robustness and/or accuracy increase

when using multimodal acquisition systems.

However, it also increases the information processing

required, and the complexity of integrating all sensors

into the same hybrid acquisition system.

It is also expected that in unimodal FMG systems,

the number of sensors will strongly influence

accuracy as they enable higher spatial resolution and

the extraction of a greater number of features

(Grushko, 2020). However, there are still several

shortcomings that need to be addressed in order to be

able to use FMG technology in commercial bionic

prostheses (Xiao,2017) (Jiang, 2020) (Xiao, 2019).

In this paper, we will describe the implementation

of a bimodal EMG/FMG system using the

physiological signal acquisition platform, BITalino

(Plux Biosignals), to make the acquisition of these

two different signals from BITalino EMG modules

and Flexiforce™ sensors from Tekscan™. The

simultaneous acquisition of EMG and FMG data was

then performed, using BITalino and OpenSignals, as

well as the optimization of the MATLAB routines for

signal processing and onset/offset detection of the ac-

quired signals, implemented in previous works within

the scope of this same project (Rodrigues, 2022)

(Rodrigues, 2023). These steps are crucial for the

extraction of features, and subsequent use of these

features to train and compare Machine Learning

models in gesture recognition, through MATLAB's

Classification Learner tool. So our main

differentiating mark is the choice of low-cost

hardware, in order to obtain a similar or even greater

efficacy with a smaller number of sensors than that

described in the literature, based on an in-depth study

that allows the selection of a smaller set of the best

characteristics and supported by an optimized

classification method. Preliminary results point to

significant gains in the effectiveness of the

classification of gestures, in line with the conclusions

of other studies (Esposito, 2018) (Rafiee, 2011).

These results, although still very preliminary, are also

better than those reported in the literature for

commercial systems with EMG sensors, with an

accuracy of 87.96% vs 84.60% for these systems

(Jiang, 2017).

2 MATERIALS AND METHODS

This work involved the selection of EMG and FMG

sensors as well as the platform for robust data

acquisition. Subsequently, it was necessary to

implement the filters for signal processing, namely

the EMG signal, as well as for the detection of

onset/offset. Finally, the features of the EMG and

FMG signals to be extracted were selected and the

entire methodology for the application of the

classifiers was developed. The main objective is to

analyze the improvement in efficacy achieved with

this bimodal system but also to optimize the

application of these classifiers.

2.1 EMG and FMG Signals

The EMG signal is a widely used tool in the detection

of motion intent in commercial bionic prosthetic

applications. However, the search for additional

information on muscle activity has motivated the

exploration of complementary techniques, such as

force myography (FMG).

The EMG signal is the electrical expression of

muscle activity, in this case captured by surface

Gesture Recognition Through the Implementation of a Bimodal Acquisition System Using EMG and FMG Signals

1019

electrodes placed on the skin on the study muscle.

The amplitude of the EMG signal, which is stochastic

(random) in nature, is influenced by the strength of

muscle contraction and usually ranges from 0 to 10

mV peak-to-peak, or from 0 to 1.5 mV RMS. The

EMG signal is particularly useful in the 0-500 Hz

frequency range, with the dominant energy in the 50-

150 Hz range. This signal characteristic is illustrated

in Figure 1, which shows power density spectra of

EMG signals from different hand gestures.

Figure 1: Power density spectra of EMG signals in hand

gestures (from (Xiao, 2019)).

FMG is a non-invasive technique that makes use

of pressure sensors placed on the skin above the

muscles to capture changes in pressure and volume

associated with the activation and deactivation of

superficial muscle groups. Instead of measuring

muscle electrical activity like EMG, FMG records

mechanical changes, thus capturing distinct

information, which can be valuable in the context of

bionic prostheses.

Although FMG has benefits such as robustness to

changes in skin impedance and sweating, and less

sensitivity to sensor positioning, it faces challenges

such as sensitivity to unintentional movements and

external interference. These limitations could be

addressed through the project of a novel 3D printed

adapter that achieves a more solid fixation of the

sensor as well as the study of filtering techniques that

would be able to cancel the noise induced by these

sources.

For the acquisition of physiological signals, we

used BITalino (r)evolution. This platform is

distinguished by its ability to integrate a wide diversity

of sensors as electromyography (EMG),

electrocardiography (ECG), accelerometer (ACC) and

many others.

In the context of this work, the BITalino board was

used to collect EMG and FMG signals. The EMG

signals were obtained using two BITalino's own EMG

sensors. On the other hand, the capture of FMG signals

required the use of two external FSR 402 sensors,

which, after a signal conditioning circuit, were

integrated into BITalino. Table 1 summarizes the main

technical specifications of BITalino (r)evolution.

Table 1: BITalino (r)evolution: technical specifications.

Sampling Rate 1, 10, 100 ou 1000 Hz

Analog Inputs

4 in (A1-A4, 10-bit) + 2 in (A5-A6, 6-bit) + 1 out (8-bit)

Digital Inputs

2 in (1-bit) + 2 out (1-bit)

Connectivity Bluetooth Class II v2.0 (range till 10 m)

Figure 2: Previously stored data file in Opensignals, showing EMG (top) and FMG (bottom) signals for the "hand opening"

gesture.

WHC 2025 - Special Session on Wearable HealthCare

1020

Table 2: Technical specifications of the EMG sensor.

Gain 1009

Ran

e ±1.64 mV

(

com VCC = 3.3 V

)

Bandwidth 25-480 Hz

Power Volta

e 2.0-3.5 V

ut Im

edance 7.5 GΩ

CMRR 86 dB

As previously mentioned, the monitoring of the

electrical activity of the flexor and extensor muscle

groups of the forearm was done using two BITalino

EMG sensors, specially designed for sEMG

acquisitions. It is compatible with gel and dry

electrodes, and offers high-quality data with low

noise due to its bipolar configuration. The EMG

sensor is responsible for analog filtering,

amplification, and A/D conversion of the signal.

Table 2 presents the technical specifications of

BITalino's EMG sensor.

Within the scope of this project, sensors of the

FSR 402 model were selected. Two of these sensors

were applied, one for each muscle group under study:

the flexor and extensor forearm. The choice of FSR

sensors is justified by their ability to detect variations

in force from an initial/resting state, rather than

providing an accurate measurement of the applied

force. This property is essential for FMG systems in

gesture recognition, where the goal is not necessarily

to quantify the exact force being applied, but to

identify if there is any force being applied and how

that force changes over time. The FSR 402, in

particular, was chosen for its active area (14.7 mm

diameter) and minimum actuation force (0.1 N),

which were considered suitable for the application in

question.

2.2 Data Acquisition

EMG and FMG signals were collected

simultaneously from each participant, using the

BITalino platform with four acquisition channels:

two for EMG and two for FMG. Data acquisition

from these four channels is commanded by the

microcontroller unit of BITalino according to the

previously defined acquisition rate. One pair of

EMG/FMG sensors was placed in the extensor

muscle group of the forearm and the other in the

flexor muscle group.

BITalino transmits the data via Bluetooth to a PC,

where the data that is being acquired it is visualized

in real-time and stored for further processing using

OpenSignals software. Participants were instructed to

perform five gestures: open, close, pinch, point, and

thumb-up. Each collected data file contains

approximately ten activations of each gesture.

The implementation of signal acquisition went

through the following steps:

1. For each acquisition session, EMG sensors

(in bipolar configuration) were positioned in the

flexor and extensor muscle groups, with a separation

of approximately 2 cm;

2. Between the two active electrodes, an FSR

sensor (on a rigid PVC base) was fixed with an

adhesive;

3. A velcro tape was applied to the forearm

over the two FMG sensors simultaneously to stabilize

the sensors in place;

4. Each participant was instructed to perform

a series of activations of a specific type of gesture,

with durations and rest intervals between activations

ranging from 1 to 3 seconds, to ensure the

representativeness of the data collected. During data

collection, the participant was asked to remain as

relaxed as possible between activations and to keep

the elbow joint still, to minimize the influence of

residual muscle strains on the collected data;

5. Each series of activations was recorded in a

separate file with the name of the gesture performed,

using the OpenSignals software. The sampling rate

was 1000 Hz. Figure 3 shows images of signal

acquisition.

Figure 3: Acquisition of EMG and FMG signals: a) On the

clasp of the hand; b) Opening the hand.

2.3 Data Processing

As illustrated in Figure 4, the EMG and FMG signals

are then initially acquired by BITalino, where they

undergo basic preprocessing, which includes

Gesture Recognition Through the Implementation of a Bimodal Acquisition System Using EMG and FMG Signals

1021

amplification and analog filtering as it is the case of a

low-pass filter to cancel high-frequency noise

(>500 Hz).

Figure 4: Steps of EMG and FMG signal processing.

After preprocessing, the data enters the phase of

extracting the characteristics of the most relevant

signals for the discrimination of gestures. Previously,

it is necessary to detect signal onsets and offsets in

order to identify the periods of muscle activation.

The signals are then forwarded for offline

processing in MATLAB. Here, additional denoising

and bandpass filtering operations are performed to

maintain only the relevant frequencies. The signal

offset is also removed.

Using the MATLAB software, the signals are

processed and their features are extracted, through a

set of previously developed routines [1,14]. This set

comprises a main routine, with the pipeline, along

with auxiliary functions for onset/offset detection and

feature extraction from EMG and FMG signals.

The main routine, implemented in MATLAB,

performs a series of critical steps in signal processing:

1. EMG signal filtering: For each text file

(with EMG and FMG data), the code applies a

bandpass filter from 20 to 500 Hz to the EMG signals;

2. Wavelet Denoising: EMG signals go

through a second stage of noise reduction, this time

using the wdenoise function of MATLAB's Wavelet

Toolbox. This technique, which acts in the time-

frequency domain, eliminates random noises that

could be mistaken for true muscle activity;

3. Onset and offset detection of muscle

activity: this is a crucial step. The code uses the

onsetting function to determine when the muscle

actually started to contract (onset) and when it

stopped (offset). The result is time series (vectors) of

onsets and offsets of muscle contraction. The

onset/offset function is responsible for identifying the

moments when the EMG signal demonstrates

significant activity. The function does this by full-

wave rectification of the signal, applying a moving

average to calcu-late the test function, and setting a

threshold for onset detection. If the signal falls be-low

this threshold, an offset is detected. In addition, the

function also ensures that the detected activity

moments have a minimum duration to avoid false

detections (650 ms);

4. Corresponding activations: the code looks

for muscle activations that coin-cide between the

EMG signals of the two muscle windows (extensor

and flexor). The onset and offset times of the FMG

signals are given by the values saved for the

corresponding EMG signals. The tolerance for

coincidence is given by the value of the constant

tolerance_window, and has been maintained at 500

ms. Figure 5 shows an example of the signals

acquired with the detection of the onsets and offsets

of each muscle activation.;

5. Feature extraction: For each muscle

activation that matches, the code extracts a set of

features from both the EMG and FMG signals.

Features are measures that provide a deeper

understanding of the signals, which would otherwise

be very difficult to interpret.

The extract_emg_features and extract_fmg_features

functions were used to extract characteristics from the

EMG and FMG signals, respectively. These functions

compute a set of characteristics, both in the time and

frequency domains (in the case of EMG), for each

instance of a gesture. In total, thirty-six characteristics

were extracted, twelve EMG and six FMG for each

muscle group.

Finally, each feature vector is labeled with the

corresponding gesture (which appears in the data file

name) and the data is prepared for classification. This

data is then used to train a classification model, which

identifies gestures based on the characteristics

extracted from the signals (Pires, 2023).

3 RESULTS

The preliminary results of this study show significant

improvement of efficacy on gesture recognition using

a bimodal EMG/FMG acquisition system. This is

accomplished from a detailed study of the application

of different machine learning models.

WHC 2025 - Special Session on Wearable HealthCare

1022

Figure 5: Example of FMG raw-signals and EMG raw and pre-processed signals. Green and red vertical lines shown are,

respectively, the onset and offset time for each muscle activation.

3.1 Dataset

In this study, three healthy individuals participated,

and the dataset was formed from the thirty-six

characteristics of the EMG and FMG signals

extracted from each activation, in each of the files

corresponding to each of the five gestures.

In total, seventy data files suitable for the following

stages of the study were recorded, distributed as

follows: hand opening (14), hand closing (16), pinch

(11), thumbs-up (16) and pointing (15). These files

were selected after discarding others due to

acquisition problems, such as excessive noise, and

incorrect positioning and/or improper fixation of the

sensors.

Figure 6.(a) shows the dataset for each gesture, while

Figure 6.(b) shows how the total of thirty-six features

extracted from each activation are distributed. In fact

the amount of signal characteristics extracted from

extensor and flexor muscles is equal.

However, of these eighteen characteristics, only six

are extracted from the FMG signal. Of these six

characteristics, only two are different from those

extracted from the EMG signal. In Figure 6.(c) all the

characteristics are presented, showing whether they

are common to both signals or from only one of the

signals, through the use of different colors.

It is also possible to observe that there are only three

characteristics (Mean Frequency, Peak Frequency

and Mean Power Spectral Frequency), and only from

EMG signal, that are frequency domain being all the

rest time domain, which are usually preferred in

sEMG based pattern recognition as they are easy and

quick to calculate since they are based on the

amplitude of the EMG signal (Christopher et al.,

2018)

The collected data from the EMG and FMG signals

of each muscle group, that consists on the relevant

characteristics that were extracted, is used as input for

the training of the Machine Learning models, through

MATLAB's Classification Learner, in order to predict

the execution of each gesture.

3.2 Feature Selection

The preliminary results of this study show

significant advances in the development of the

gesture recognition system. In a first phase, a

preliminary comparison of the thirty-three available

classification models was made, using accuracy (or

"effectiveness") as the main metric. In this study,

these thirty-three classification models were trained

and evaluated, using the built-in algorithms of the

MATLAB Classification Learner tool. The

techniques applied ranged from more linear

approaches, such as Quadratic Discriminant, to more

sophisticated methods, including SVMs and Neural

Network architectures. From these study six

classification models can be highlighted: Linear

Discriminant, Quadratic SVM, Cubic SVM and three

Neural Network architectures (Narrow, Medium

and Wide). These models were trained with different

Gesture Recognition Through the Implementation of a Bimodal Acquisition System Using EMG and FMG Signals

1023

Figure 6: (a) Dataset for each gesture. (b) Distribution of the amount of features extracted per muscle and per sensor type. (c)

4 features are extracted from FMG and EMG signals simultaneously (green), 8 features from EMG signal (red) and 2 features

from FMG signal (black).

Figure 7: Confusion matrices for the Wide Neural Network model. (a) validation (b) test.

feature selection methods - ANOVA, ReliefF and

Kruskal Wallis - and varying the percentage of

selected features (75, 50 or 25%). standing out with

100% of the features, achieving validation and test

accuracies of 91.7% and 93.8%, respectively. On the

other hand, the classifiers based on neural networks

showed a greater variability in their results, indicating

a sensitivity to the selection of features. In particular,

the wide neural network showed excellent

performance without feature selection, achieving

validation and testing accuracies of 95.1% and

93.8%, respectively.

Figure 7 shows the confusion matrices for an

example of the trained models (Wide Neural

Network), to show that these matrices help to

understand how each model handles the different

classes and provide a visual understanding of the

models' performances.

(a)

(b)

(c)

ass

open

(a) (b)

WHC 2025 - Special Session on Wearable HealthCare

1024

Table 3: Comparison of the accuracy of the classifiers between the use of all data, EMG only and FMG only.

Classifier Model Validation

(EMG +FMG)

Test (EMG

+FMG)

Validation

(EMG)

Test

(EMG)

Validation

(FMG)

Test

(FMG)

Linear Discriminant

74,07 75,46 62,96 64,35 50,69 53,24

Quadratic SVM

89,35 87,96 79,28 75 65,16 68,06

Cubic SVM

89,24 87,96 79,4 79,63 70,49 70,37

Narrow NN

79,51 78,7 73,96 74,54 63,19 68,52

Medium NN

85,07 81,02 74,54 75 59,49 60,19

Wide NN

88,43 82,41 78,47 79,17 67,25 68,06

3.3 Bimodal vs EMG vs FMS Efficacy

In this section, we explore the impact of combining

EMG and FMG characteristics on the performance of

classifiers. To this end, the bimodal approach was

contrasted with the more common practice that uses

exclusively EMG characteristics. Table 3 details the

performance of the six classifiers indicated above,

when they use all characteristics, only EMG

characteristics and only FMG characteristics.

4 DISCUSSION

This paper presents preliminary results of the

implementation of a bimodal sys-tem with EMG and

FMG sensors in which two EMG+FMG pairs are

placed in the flexor and extensor muscles. A total of

thirty-six characteristics of these two acquired signals

were used for three healthy individuals, and the

dataset consisted of five different gestures. The main

objective of this study is to evaluate the benefit, in

terms of efficacy in the recognition of the gestures

performed, that is obtained by the acquisition of the

FMG signal simultaneously with the EMG signal,

because this signal when used in isolation has some

limitations that result, for example, from variations in

the impedance of the skin interface.

MATLAB's Classification Learner was used, thirty-

one classifiers were applied and a study was also

made on the possibility of reducing the number of

characteristics, which will be an important point to

reduce the processing time and consequently the

response time of the bionic hand in the execution of

gestures. For this, three different methods of selection

of the characteristics were used, with different

percentages (75%, 50% and 25%) of the total of

thirty-six characteristics.

The preliminary results presented focus on the

most used metric which is accuracy but the results are

also being analyzed with other metrics, namely, F-

score and the area under the ROC curve. It is possible

to verify how different classifiers have very different

behaviors, with those that are more effective but more

sensitive to the reduction of the number of

characteristics and others that are more immune to

this selection of characteristics.

Although this evaluation of the bimodal system is

still ongoing, the results presented here reinforce the

idea, supported by previous research, that the

combination of EMG and FMG allows to improve the

efficiency of machine learning models in gesture

recognition. So, as ultimate conclusion, this study

contributes to the field of myoelectric prostheses by

exploring the implementation and testing the

efficiency of a bimodal EMG/FMG signal acquisition

system for the control of a bionic hand.

REFERENCES

Marques, J, Ramos, S, Macedo, M.P, da Silva, H.P (2020).

Study of Mechanomyographic Alternatives to EMG

Sensors for a Low-Cost Open Source Bionic Hand. In:

Inácio, P., Duarte, A., Fazendeiro, P., Pombo, N. (eds)

5th EAI International Conference on IoT Technologies

for HealthCare. HealthyIoT 2018. EAI/Springer

Innovations in Communication and Computing.

Springer, Cham.

Silva, D, Castro, S, Macedo, M.P. and Silva, H.P. (2019).

Towards Improving the Usability of Muscle Sensing in

Open Source Bionic Hand: Mechanomyography vs.

Electromyography with Novel Electrodes”. AmiEs-

2019 - International Symposium on Ambient Intel-

ligence and Embedded Systems, 1-6.

Grushko, S., Spurný, T., & Černý, M. (2020). Control

Methods for Transradial Prostheses Based on Remnant

Muscle Activity and Its Relationship with

Proprioceptive Feedback. Sensors, 20(17).

McIntosh, J., McNeill, C., Fraser, M., Kerber, F.,

Löchtefeld, M., & Krüger, A. (2016). EMPress:

Practical Hand Gesture Classification with Wrist-

Mounted EMG and Pressure Sensing. 2332–2342.

Abboudi, R. L., Glass, C. A., Newby, N. A., Flint, J. A., &

Craelius, W. (1999). A biomimetic controller for a

Gesture Recognition Through the Implementation of a Bimodal Acquisition System Using EMG and FMG Signals

1025

multifinger prosthesis. IEEE Transactions on

Rehabilitation Engineering, 7(2), 121–129.

Christopher, S., Md Rasedul, I., Assad-Uz-Zaman, M., &

Rahman, M. (2018). A Comprehensive Study on EMG

Feature Extraction and Classifiers. Open Access

Journal of Biomedical Engineering and its

Applications, 1.

Citi, L., Vidoni, R., Menoncmenon, C., Cho, E., Chen, R.,

Merhi, L.-K., Xiao, Z., Pousett, B., & Menon, C.

(2016). Force Myography to Control Robotic Upper

Extremity Prostheses: A Feasibility Study. 4.

Kadkhodayan, A., Jiang, X., & Menon, C. (2016).

Continuous Prediction of Finger Movements Using

Force Myography. Journal of Medical and Biological

Engineering, 36(4), 594–604.

Radmand, A., Scheme, E., & Englehart, K. (2016). High-

density force myography: A possible alternative for

upper-limb prosthetic control. Journal of Rehabilitation

Research and Development, 53, 443–456.

Jiang, X., Merhi, L.-K., Xiao, G. Menon, C. (2017).

Exploration of force myography and surface

electromyography in hand gesture classification,

Medical Engineering & Physics, 41, 63–73.

Jaquier, N., Connan, M., Castellini, C., & Calinon, S.

(2017). Combining Electromyography and Tactile

Myography to Im-prove Hand and Wrist Activity

Detection in Prostheses. Technologies, 5, 64.

Nowak, M., Eiband, T., & Castellini, C. (2017). Multi-

modal myocontrol: Testing combined force- and

electromyography. IEEE ... International Conference

on Rehabilitation Robotics : [proceedings], 2017,

1364–1368.

Xiao, Z. G., & Menon, C. (2017). Performance of Forearm

FMG and sEMG for Estimating Elbow, Forearm and

Wrist Positions. Journal of Bionic Engineering, 14(2),

284–295.

Jiang, S., Gao, Q., Liu, H., & Shull, P. B. (2020). A novel,

co-located EMG-FMG-sensing wearable armband for

hand gesture recognition. Sensors and Actuators A:

Physical, 301, 111738..

Xiao, Z. G., & Menon, C. (2019). A Review of Force

Myography Research and Development. Sensors,

19(20).

Rodrigues, S. and Macedo, M.P.: Algorithm for

Onset/Offset Detection of EMG Signals for Real-time

Control of a Low-Cost Open-Source Bionic-Hand. In

Proceedings of the 15th Int. Joint Conf. on Biomedical

Engineering Systems and Technologies – WHC, 872-

878. (2022).

Rodrigues, S., Macedo, M.P. (2023). A Low-Cost Open-

Source Bionic Hand Controller: Preliminary Results

and Perspectives. In: Spinsante, S., Iadarola, G.,

Paglialonga, A., Tramarin, F. (eds) IoT Technologies

for HealthCare. HealthyIoT 2022. Lecture Notes of the

Institute for Computer Sciences, Social Informatics and

Telecommunications Engineering, vol 456. Springer,

Cham.

Esposito, D., Andreozzi, E., Fratini, A., Gargiulo, G. D.,

Savino, S., Niola, V., & Bifulco, P. (2018). A

piezoresistive sensor to measure muscle contraction

and mechanomyography. Sensors (Switzerland), 18(8).

Rafiee, J., Rafiee, M. A., Yavari, F. and Schoen, M. P.

(2011). “Feature extraction of forearm EMG signals for

prosthetics,” Expert Systems with Applications, vol.

38, no. 4, pp. 4058–4067.

Pires, N. (2023). Reconhecimento de gestos através da

implementação de sistema bimodal de aquisição de

sinais EMG e FMG [[Unpublished Biomed. Eng. BSc,

thesis]. Polytechnic Institute of Coimbra.

WHC 2025 - Special Session on Wearable HealthCare

1026