Sensory Glove and Surface EMG with Suitable Conditioning

Electronics for Extended Monitoring and Functional Hand

Assessment

Giovanni Saggio, Giancarlo Orengo and Alberto Leggieri

Department of Electronics Engineering, Tor Vergata University, via Politecnico 1, 00133, Rome, Italy

Keywords: Sensory Glove, Surface EMG, Finger Movements.

Abstract: We propose and evaluate a new method for measuring and discriminating among flexion, extension,

abduction and adduction movements of hand fingers. In particular, flex sensors allowed registering flexion-

extension movements, whereas data from multi-channel surface electromyography (sEMG) electrodes

allowed discriminating adduction-abduction movements of thumb, index and middle fingers. An electronic

interface was designed to acquire and pre-process signals feeding a Personal Computer (PC), running

indigenously made routines for data recording, visualization and storing. A novel test for repeatability and

reproducibility was also proposed and successfully adopted.

1 INTRODUCTION

Sensory glove is defined as a supporting glove

equipped with sensors aimed to measure hand

assessment. But the hand is a masterpiece of

dexterity with 27 degrees of freedom (DOFs), as

determined by a widely adopted kinematic hand

model (Lin et al., 2000). Therefore, usually, only a

subset of DOFs is considered, and a complete 27

DOFs sensory glove is rarely realized. Those 27

DOFs include flexion/extension of the fingers’ joints

(i.e. the phalanges come closer/away together),

abduction/adduction between fingers (i.e. the

movements that bring one finger away and close

from the adjacent), and rotational/translational

capabilities of the wrist.

Among all, the abduction/adduction movements

have been found to be the most difficult to measure.

This is because flexion/extension and

rotational/translational capabilities can be measured

directly with a flex sensor lying on-top of the finger

or of the wrist skin surfaces, but the same approach

is not possible to reveal abduction/adduction

capabilities.

To overcome the problem, the commonly

adopted solution is to arrange a strip rectangular flex

sensor upright the dorsal aspect of two adjacent

fingers (Figure 1), and not between the fingers, as it

would be preferable, but impracticable to avoid

grasping limitations. This upright arrangement of the

flex sensor can suffer from mechanical instability

(since possible misplacements of the sensor during

usage) and a subsequent too poor measurement

accuracy, therefore here we propose to adopt surface

electromyography (sEMG) sensors rather than flex

ones.

Figure 1: Strip rectangular flex sensor upright the dorsal

aspect of two adjacent fingers to measure

abduction/adduction angles.

In practice, we propose a combination of flex

sensors and sEMG sensors for overall functional

hand assessment, In fact, sEMG reveals all

movements, so that, subtracting flexion-extension

measures (obtained from flex sensors), we can

obtain the measure of the abduction/adduction

capabilities.

206

Saggio, G., Orengo, G. and Leggieri, A.

Sensory Glove and Surface EMG with Suitable Conditioning Electronics for Extended Monitoring and Functional Hand Assessment.

DOI: 10.5220/0005704702060213

In Proceedings of the 9th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2016) - Volume 4: BIOSIGNALS, pages 206-213

ISBN: 978-989-758-170-0

For simplification purpose, our attention was

limited to the first three finger movements, which

are necessary to accomplish the most important tasks

of human hand. The sEMG signals were collected by

four couples of differential electrodes positioned on

the right hand.

Data coming from both flex and sEMG sensors

fed an Arduino microcontroller board connected to a

PC. Ad-hoc PC running software was developed to

represent the finger posture.

The sensory system is described in the second

section of the paper, the electronic interface in the

third section, the procedures used during the

measurement sessions with their results in the fourth

section, discussion and conclusions in the last

section.

2 MEASUREMENT SYSTEM

2.1 Sensory Glove

Sensory gloves have been gaining more and more

relevance in literature, even if the cost, the

calibration procedure and the need of different size

(to fit different hands) have limited their diffusion in

clinical protocols (Gentner and Classen, 2009).

The sensory glove (Figure 2) used in this work

has characteristics already reported (Saggio et al.,

2012). It is equipped with 14 flex sensors (Figure 3,

by Flexpoint Sensor Systems Inc. South Draper UT,

USA), already reviewed in (Saggio et al., 2015), for

flex-extension of all the fingers’ joints, but we

limited our attention to only the three sensors on the

metacarpo-phalangeal joints of thumb, index and

middle finger.

Figure 2: Lycra-based glove. Signals from only three flex

sensor were used in this work.

Figure 3: Flex sensor from Flexpoint.

2.2 Surface EMG

Signal acquisition from sEMG sensors is rather

simple, but decoding data can be challenging,

because signal parameters (e.g. amplitude,

frequency) are not easily linkable to subcutaneous

effects. For example, the mean frequency is

generally related to the muscle fatigue, even if it is

not yet clear if the shift toward low frequencies is

related to the speed decrease of muscular fibers

(Merletti et al., 1990) rather than to the employment

of slower motor units (Rau et al., 2004). The

measure of muscular strength is often associated

with the signal amplitude, even if it also depends on

the electrode position and size with respect to the

muscle, as well as the distance between electrodes.

Moreover, cross-talk between different muscles adds

noise to signal parameters measurement (Rau et al.,

2004).

Figure 4: Location of 4 the couples of differential sEMG

electrodes. The two couples of electrodes to register the

index FDI, the middle TDI and the reference electrode on

the wrist are shown on the leftmost photo, whereas the two

couples for the thumb TA and STA on the rightmost. The

cuff increases the sensor adherence to skin.

The use of sEMG to measure finger movements

is mostly applied to prosthesis control by amputated

people (Gentner and Classen, 2009; Riillo et al.,

2014), positioning electrodes on the forearm and

getting sEMG signals from extrinsic hand muscles.

On the other hand, discrimination of finger

movements by mean of sEMG signals associated

with intrinsic muscle fatigue is still an unexplored

research field, so much that even international

recommendations have not yet provided instructions

about electrode positioning for this kind of muscles.

As a novelty, sEMG electrodes were positioned on

the hand, in this case, to allow the measurement of

abduction/adduction fingers’ movements.

The sEMG electrodes used in this application are

the “3M Red Dot” with circular shape and 60mm

Sensory Glove and Surface EMG with Suitable Conditioning Electronics for Extended Monitoring and Functional Hand Assessment

207

diameter, their position as in Figure 4. sEMG sensor

couples, for differential measures, were attached to

skin surface, each along the center of the relative

muscular fiber: the first dorsal interosseus (FDI,

which controls the index adduction) and the third

dorsal interosseus (TDI, which controls the middle

adduction), on the hand span, and the short thumb

abductor (STA) and the thumb adductor (TA), on the

hand back.

The reference electrode on the wrist provides a

common reference to the patient and to an adopted

differential amplifier, to drive the isolated patient to

a positive voltage respect to ground, and to reduce

the 50 Hz interference from the power line. This

electrode needs to be located on a neutral region as a

bond, not affected by the voltages created by

metabolic activity. According to SENIAM

recommendations (Stegeman and Hermens, 2007), it

was located on the carp pisiform bond.

3 ELECTRONIC INTERFACE

The electronics interfaced sensors with PC. It based

on analog amplifiers connected to the analog inputs

of an Arduino-based microcontroller device (by

Smart Projects, Strambino, Torino, Italy) (Figure 5),

5V supplied via PC USB.

Figure 5: Stripboard of the electronic interface.

3.1 Flex Sensor Interface

Flex sensors are capable to change their resistance

when bent. The output resistance of each sensor was

converted into voltage values by means of voltage

dividers (Figure 6), whose fixed resistances were

calculated in (Saggio et al., 2012). After a buffering

stage with an operational amplifier (OA), the voltage

values are connected to the analog input A4, A5 and

A6 of Arduino for the three fingers, respectively.

Figure 6: Schematic of the glove sensors electronic

interface.

3.2 SEMG Interface

The first stage of a biopotential amplifier is an

instrument differential amplifier (INA) with high

CMRR. Electrical interference induced from the

power line, or originating from other sources of

biopotentials in more remote parts of the body, are

detected simultaneously by both electrodes and were

rejected by the first stage of the INA as common-

mode signals. The chosen INA was the INA114 (by

Burr Brown, Tucson, Arizona, USA), which features

single supply mode and very high input impedance.

It is dc-coupled to the electrodes via current-limiting

resistors and fault current limiters.The ground path

for the input bias current is therefore the body itself.

Although the INA114 has 115dB of CMRR, it

decreases at a rate of 20dB/dec, becoming too low to

reject strong common-mode RF signals beyond

100kHz. These random interference signals can also

generate dc offset through RF rectification, which,

amplified by the following gain stages, causes

response errors and even saturation of solid-state

devices.Therefore, the voltage gain of the INA was

adjusted to 12 only, planning to yield the required

gain with the following stage.

The body induced voltage from 50Hz main

causes a displacement current through the patient,

resulting in a common-mode voltage between the

two recording electrodes and theamplifier common.

This can be reduced twisting and shielding the

electrode leads. Figure 7 shows a model of all the

coupling capacitances and their calculated values

(Neuman, 1978). Moreover, asymmetry in the two

electrode impedances, due to random contact

variations, transform a common-mode voltage into a

differential one. Since the electrode impedance

cannot be enough low, the noise signal is normally

higher than the sEMG signal, which is in the range

0.1-1mV (Neuman, 1978). The common practice is

to add a further electrode on the patient. But instead

to connect this electrode to the amplifier reference

voltage, causing a dispersion current to arise and a

voltage to drop on the electrode resistance (useful to





BIAS

Arduino

sen

ref

out



in1



in2



in3



inN



BIOSIGNALS 2016 - 9th International Conference on Bio-inspired Systems and Signal Processing

208

difference the two reference voltages) (Neuman,

1978), it was connected to the driven-right-leg

circuit (DRL) in a feedback loop. This is because in

ECG systems it is effectively connected to the

patient right leg through another electrode, whereas

in sEMG systems is connected to the wrist,

according to SENIAM recommendations (Stegeman

and Hermens, 2007). In this way, the common-

voltage is attenuated by the amplifier gain with

respect to a direct connection.

Figure 7: Circuit schematic of the parasitic coupling

capacitances and the driven-right-leg circuit connecting

the patient to the amplifier reference voltage to eliminate

the 50 Hz noise.

In order to eliminate the so called “baseline

noise” (a slow oscillation of the average signal

value) which, together with movement artefacts, is

considered a main noise source from electrodes,

added to RF rectification and 1/f noise originated by

the INA, a dc suppression circuit (Figure 8) were

inserted after the INA amplifier.

Figure 8: Output dc suppression circuit.

The integrator in the feedback loop provides ac-

coupling with the following amplifier, thus changing

the signal baseline with its reference voltage

(Spinelli et al., 2003). The cutoff frequency was

chosen, according to suggestions of International

bodies (Stegeman and Hermens, 2007), equal to

10Hz.

The INA114 is followed by the second

amplifying stage, a non-inverting low-pass

amplifier, configured to obtain 460Hz bandwidth

and a voltage gain of 180. It employs a

TLC25L4ACN operational amplifier (by Texas

Instrument, Texas, USA), which operated in single

5V supply mode.

Finally, a dc restoration circuit, composed of a

buffered RC high-pass filter, with a cutoff frequency

of 10Hz, is used to guarantee a dc level equal to

2.5V, as required by the Arduino analog-to-digital

converter (ADC).

The power spectral density (PSD) of EMG

signals is within 5-500Hz, but for sEMG is allowed

even less bandwidth. The overall performance of the

band-pass analog interface is then a 2160 voltage

gain with 12Hz and 460Hz cutoff frequencies, given

a 1kHz sampling frequency of the ADC.

The frequency response of the second stage

amplifier, however, features only 20dB/dec of

selectivity, which is too low for an anti-aliasing

filter. For this reason, a Butterworth filter was

realized, based on a second order Sallen-Key cell

with 40dB/dec attenuation (Figure 9). The cutoff

frequency of 150Hz was to provide 20dB attenuation

@ 470Hz (a higher degree filter would provide more

attenuation but less smooth time response and

unacceptable transients). The full schematic of the

sEMG interface is shown in Figure 10.

Figure 9: Second order anti-aliasing Butterworth filter.

Figure 10: Schematic of the 4-channel electronic interface

for sEMG sensors.

Sensory Glove and Surface EMG with Suitable Conditioning Electronics for Extended Monitoring and Functional Hand Assessment

209

3.3 Digital Interface

For signal conditioning, signal dc offset was set to

2.5V, which was also chosen as the patient reference

voltage, given through the reference electrode.

Therefore, this voltage represents the signal ground

for the circuit.

Data were conditioned by an Arduino Micro

board, based on an ATmega32u4 (by Atmel

Corporation, San Josè, California, USA)

microcontroller device.

In order to maximize the communication speed,

data were stored in a 14 byte register (2 byte each

10-bit value), and send as binary data (Serial.write()

command) without ASCII conversion (Serial.print()

command) every one millisecond. Arduino Micro

sends its data to the computer at a speed of 400kbps.

A Matlab application was developed to record

and save the data in text files, setting the baud rate to

460800 bps, the buffer length, and specifying the

data length (two bytes always positive) according to

the script uint16. The code reads 100 records at the

same time, each corresponding to a 10-bit digitalized

voltage printed by Arduino on the serial port, and

plots them in real time. The effective voltage is

obtained from its digital value  from the equation

V=5N/1023.

Further noise was detected on the resulting

signal, which was filtered through a digital band-

pass Butterworth filter, with a low cutoff frequency

of 20Hz, and a high cutoff frequency of 495Hz, to

attenuate the high frequency harmonics generated by

the sampling process. The root mean square (RMS)

value is then calculated on a window of 300

samples, shifting it by 75 samples each time.

4 sEMG MEASUREMENTS

In this section data registration and modeling of

sEMG static measurements of abduction/adduction

posture of the thumb, index and middle fingers are

presented. Data were acquired from six able-bodied

subjects, 3 male (M1,M2,M3) and 3 female

(F1,F2,F3), five right-hand and one left-hand, each

one using his/her dominant hand.

Measurement results from different subjects

show a remarkable spread. Then assessment of

sEMG activity needs to be each time calibrated on

the subject. A personal characterization session was

defined to this purpose, where the number of

measurements were reduced as much as a provided

tolerance is still guaranteed by the extracted model.

At this point, the measurement session related to a

particular task can be start.

4.1 Wise Test

Since the novelty of our approach, a new test to

evaluate the repeatability and reproducibility of

finger abduction/adduction movement assessment

was created. It was based on the Wise test provided

for flexion/extension measurement, used to evaluate

the performance of the electronic gloves (Gentner

and Classen, 2009; Dipietro et al., 2003).

It consisted in placing and re-placing the hand in

known postures always with the glove and sEMG

sensors donned, to evaluate measurement

repeatability, and placing and re-placing the hand in

known postures after donning and doffing the

sensors, to evaluate the measurement

reproducibility. In particular, the postures were a)

flat hand with closed fingers (starting posture), b)

flat hand with 20° thumb abduction, c) flat hand

with 10° index abduction, d) flat hand with 10°

middle abduction.

A further posture with the maximal voluntary

contraction (MVC) is also drawn for each finger.

The three abduction angles to be measured are far

from the MVC reported in Table 1, then easily

performed and repeated by each subject (Merletti et

al., 1990), which had to open the fingers up to the

chosen abduction angle and hold it for 2s, during

which the sEMG signals are registered, then back to

the starting posture, where the resting sEMG signals

are recorded. This task was repeated 10 times with a

rest interval of 10s between them. After this

sequence, the subject was asked to perform

abduction to the MVC for each finger, in order to

identify a regression of the sEMG signal intensity

against the abduction angle with three points, that is

0°, 10° and MVC. Finally a data block is created.

This procedure is repeated 10 times with a resting

period of 3min each time. In order to evaluate the

repeatability, the same sEMG sensors were used to

measure the two positions (0°-10°), performing task

A-C, whereas the sEMG sensor were changed after

each sequence to evaluate the reproducibility (task

B-D).

Table 1: MCV values for index and middle abduction.

subject M1 M2 M3 F1 F2 F3

index 25 25 30 30 15 20

middle 25 25 30 30 25 20

Each block is composed of 2000 elements, obtained

from the RMS value of the signal samples registered

BIOSIGNALS 2016 - 9th International Conference on Bio-inspired Systems and Signal Processing

210

during the resting intervals. The first 25 samples

were eliminated, because affected by movement

artefacts, and the RMS value is calculated on the

next 150 samples.

From the total mean ̅



of the 10 blocks of the

test, and the standard deviation  of all the blocks of

measurements, the uncertainty of measurements can

be expressed as ̅



 2

⁄

, from which the test

results were measured through the normalized

standard deviation (percentage) , that is the ratio

between standard deviation and the total mean value.

It ranges from 8.28%, corresponding to 10° 

0.414° (FDI-testA-F2), to 21.1% (FDI-testB-F1),

corresponding 10°  1.06°. The results for the

middle finger are 0.6° (TDI-testA) and 1.15°

(TDI-testB). Results for test B (0° abduction)

without sEMG signals are generally worst, because

the baseline measurement is more affected by noise.

4.2 Regression Models

To characterize the abduction/adduction angles

against the RMS value of the sEMG signal, more

data points are needed than two used for the Wise

test. For this purpose, a hinge with a fixed and a

mobile arm was inserted between the couple of

fingers under measurement, in order to provide a

finger posture with the right abduction angles,

ranging from 0° to the MVC, with a 5° step. The

mobile arm of the hinge was moved by a Trinamic

step motor driven by a Labview interface. Although

the hinge could be seen as a constrain, the subject

was asked to provide the same strength as if the

hinge would not be used.

Although regression models should allow

obtaining the abduction/adduction angle from the

measured voltage, as independent variable, it would

be hard to yield models from not always univocal

relations. On the other hand, if the independent

variable is the angle step, as in this case, good fitting

results can be yield. The inverse function can be

obtained through a numerical algorithm, because the

regression functions are hardly ever invertible.

Different regression models were tried to

interpolate the available data, either polynomial

models from the linear one to fifth degree, and the

monomial/binomial exponential models, which are

described in Equations (1) and (2). Each model was

statistically evaluated by mean of the 



coefficient,

which computes the correlation degree between data

and model points.





(1)









(2)

4.2.1 Index/Middle Abduction/Adduction

The measurement of the sEMG voltage, representing

the muscle fatigue, against the imposed angle is

reported in Figures 11 and 12 for the FDI (index)

and the TDI (middle), respectively, here limited to

subject M1 for sake of brevity.

Figure 11: FDI sEMG assessment for subject M1 of

middle abduction (red circle) and adduction (blu square)

with standard deviation segments, superimposed to the

corresponding binomial exponential regression models

(continuous for abduction and dashed for adduction).

Figure 12: TDI sEMG assessment for subject M1 of

middle abduction (red circle) and adduction (blu square)

with standard deviation segments, superimposed to the

corresponding binomial exponential regression models

(continuous for abduction and dashed for adduction).

The measurement of the abduction and the

adduction movements are represented with different

symbols. The difference between the muscle fatigues

in the two phases is marked more for the FDI than

TDI in all subjects. Figures also report the measure

Sensory Glove and Surface EMG with Suitable Conditioning Electronics for Extended Monitoring and Functional Hand Assessment

211

of the standard deviation, whose results are

summarized in Table 2, as mean value between the

six subjects, and compared with the corresponding

ones from the test.

Taking into account the regression R

parameter

reported in Table 3, the linear model demonstrated

to be not suitable to represent the voltage/angle

relation. The performance of the binomial

exponential regression model is also plotted in each

figure.

Table 2: Comparison of the mean normalized standard

deviation between subjects.

FDI TDI

Wise test 10.88% 11.17%

Hinge meas 10.87% 10.56%

Table 3: Comparison of 



correlation coefficient for five

polynomial and two exponential regression models of

index FDI and middle TDI measurements.

FDI TDI

model abduct adduct abduct adduct

0.91 0.71 0.76 0.79

1 0.95 0.96 0.95

1 0.99 1 1

1 1 1 1

mon. 0.99 0.99 0.90 0.92

bin. 0.99 0.99 0.99 0.99

4.2.2 Thumb Abductor/Adductor

The same procedure was used to measure the radial

abduction/adduction of the thumb, except for, this

time, both the thumb abductor (TA) and the short

thumb adductor (STA) operate an abduction

movement when their intensity is growing, and an

adduction movement when is decreasing.

Table 4: Comparison of 



correlation coefficient for five

polynomial and two exponential regression models of TA

and STA adduction/abduction measurements.

TA STA

model abduct adduct abduct adduct

0.76 0.62 0.65 0.77

0.98 0.94 0.98 0.99

1 0.99 1 1

1 1 1 1

mon. 0.98 0.91 0.87 0.96

bin. 1 0.98 1 1

Measurement results and the binomial

exponential regression models are reported in

Figures 13 and 14, whereas the 



results of each

model are reported in Table 4.

Figure 13: TA sEMG assessment for subject M1 of

abduction (red circle) and adduction (blu square) with

standard deviation segments, superimposed to the

corresponding binomial exponential regression models

(continuous for abduction and dashed for adduction).

Figure 14: STA sEMG assessment for subject M1 of

abduction (red circle) and adduction (blu square) with

standard deviation segments, superimposed to the

corresponding binomial exponential regression models

(continuous for abduction and dashed for adduction).

In this case, it is worth noting that there is

ambiguity for high abduction/adduction angles for

TA and low angles for STA, suggesting that the two

measurements can be complementary within the

same algorithm, provided to discriminate the thumb

position.

4.3 Finger Position Recognition

SEMG measurements of abduction/adduction were

integrated with those of flexion/extension from flex

sensors, to provide a complete identification of the

BIOSIGNALS 2016 - 9th International Conference on Bio-inspired Systems and Signal Processing

212

finger position. Bend angles were obtained through a

linear relation with the sensor response.

The intrinsic muscles involved in the abduction

movements, however, are activated also during

flexion/extension movements. In order to

discriminate the sEMG signal corresponding to the

abduction movement, the flex sensor response was

used.

Taking into account that for the index and middle

abduction the maximum amplitude is obtained

without any finger flexion, on one hand, and the

abduction angle is constrained to zero at the

maximum finger flexion, on the other, it was defined

a bend coefficient  to modulate the sMEG

response, according to equations (3) and (4)

  90°  



 90°

⁄

(3)







∙





(4)

On the contrary, the flexion and abduction

movements of the thumb are independent, and can

be actuated simultaneously.

5 CONCLUSIONS

In this paper the combination of flex and sEMG

sensors was considered to measure

abduction/adduction capabilities of the first three

fingers. The sEMG signals were integrated with

information taken by flex sensor, to discriminate the

muscle fatigue devoted to abduction/adduction

movements from that devoted to flexions/extensions.

As a novelty, sEMG electrodes positioned on the

hand allowed a measurement of abduction/adduction

fingers’ movements.

Data from both sensors’ type fed an ad-hoc

realized circuitry based on an Arduino

microcontroller. PC running software was developed

to represent the finger posture with bar plots.

A problem which is still to overcome is the

personal characterization of the system, which has to

be accomplished by each subject before to start the

measurement session. Moreover, in the case of

thumb position recognition, both TA and STA

measurements need to be simultaneously available

to extract the actual abduction/adduction angle.

Future developments can be the reduction of

cross-talk between sEMG sensors, the integration of

the thumb opposition measurement, and the dynamic

posture recognition other than the static ones.

ACKNOWLEDGEMENTS

This paper was partially based on a work supported

by the Italian Space Agency (ASI), contract #2013-

081-R0, for which we would like to thank Prof.

Mariano Bizzarri, Dr. Simona Zoffoli and Dr.

Francesca Ferranti.

REFERENCES

Dipietro L., Sabatini A., Dario P., 2003. Evaluation of an

instrumented glove for hand-movement acquisition.

Journal of Rehabilitation Research and Development,

N. 2, 181-191.

Gentner R., Classen J., 2009. Development and evaluation

of a low-cost sensor glove for assessment of human

finger movements in neurophysiological settings. J.

Neurosci. Methods, 178, 138-147.

Lin J., Wu Y. and Huang T. S., 2000. Modeling the

constraints of human hand motion. IEEE Workshop on

Human Motion, 121-126.

Merletti R., Knaflitz M., & De Luca C. J., 1990.

Myoelectric manifestations of fatigue in voluntary and

electrically elicited contractions. J. Appl. Physiol.,

69(5), 1810-1820.

Neuman M. R., 1978. Biopotential amplifiers”, Medical

Instrumentation: Application and Design. Webster J.

G., Ed. Boston, MA: Houghton Mifflin.

Rau G., Schulte E., & Disselhorst-Klug C., 2004. From

cell to movement: to what answers does EMG really

contribute?. Journal of Electromyography and

Kinesiology, 14(5), 611-617.

Riillo F., Quitadamo L. R., Cavrini F., Gruppioni E.,

Pinto C.A., Pastò N.C., Sbernini L., Albero L., Saggio

G. 2014. Optimization of EMG-based hand gesture

recognition: supervised vs. unsupervised data

preprocessing on healthy subjects and transradial

amputees. Biomedical Signal Processing and Control,

Vol. 14, 117-125.

Saggio G., Lagati A., Orengo G., 2012. Wireless Sensory

Glove System developed for advanced Human

Computer Interface. International Journal of

Information Science, 2(5), 54-59.

Saggio G., Riillo F., Sbernini L., Quitadamo LR, 2015.

Resistive flex sensors: a survey. Smart Materials and

Structures, Volume 25, Number 1, pp. 1-30.

Spinelli E. M., Pallàs-Areny R., & Mayosky M. A., 2003.

AC-coupled front-end for biopotential measurements.

IEEE Trans. on Biomedical Engineering, 50(3), 391-

395.

Stegeman D., & Hermens H., 2007. Standards for surface

electromyography: The European project Surface

EMG for non-invasive assessment of muscles

(SENIAM).

Sensory Glove and Surface EMG with Suitable Conditioning Electronics for Extended Monitoring and Functional Hand Assessment

213