STUDYING REPEATABILITY

In this section we will briefly present our experiments.

our experiments on Repeatability 

The aim of our experiments was to study repeatability for electromyographic signals in order to show how postures classification changes in relation to different conditions. At first, we studied repeatability during a short period, essentially during a whole day. Secondly we analyzed the repeatability issue during a long term period, i.e 4 days. Our last experiment, but not least, consisted in introducing slight changes on the classification tool configurations.

For this purpose, we used the NinaPro acquisition datasets and experimental materials. Further details on the NinaPro project can be found in [3, 4].

As for tools configuration, we sticked to all the successful sEMG classification methods showed in [1, 2] and in many other researches on this topic, that are widely reported in the following sources:

1. Kuzborskij, Ilja, Arjan Gijsberts, and Barbara Caputo. “On the challenge of classifying 52 hand movements from surface electromyography.” Engineering in Medicine and Biology Society (EMBC), 2012 Annual International Conference of the IEEE. IEEE, 2012.
2. Atzori, M.; Gijsberts, A.; Heynen, S.; Hager, A.M.; Deriaz, O.; van der Smagt, P.; Castellini, C.; Caputo, B.; Muller, H., “Building the Ninapro database: A resource for the biorobotics community,” in Biomedical Robotics and Biomechatronics (BioRob), 2012 4th IEEE RAS & EMBS International Conference on , vol., no., pp.1258-1265, 24-27 June 2012 presented at Proceedings of the Fourth IEEE RAS/EMBS International Conference on Biomedical Robotics and Biomechatronics
3. https://www.idiap.ch/project/ninapro
4. http://publications.hevs.ch/index.php/authors/show/797

MATERIALS

In agreement with our prof. B.Caputo and her highly qualified team, we decided to create this page in order to share the experimental procedure.

In this section we will show each stage of the preprocessing data and features extraction methods. 

PreProcessing

Raw data need preprocessing steps in order to be used in a successful way. We did synchronization, relabelling, filtering and windowing on the raw signals.

Each step can be implemented through MATLAB interface. For this reason, we strongly recommend visiting: http://it.mathworks.com/

Downloadable source codes with instructions on tools configurations can be found on this link on GoogleDrive platform:

Materials.zip

In order to reproduce each step in a proper way we suggest reading

Kuzborskij, Ilja, Arjan Gijsberts, and Barbara Caputo. “On the challenge of classifying 52 hand movements from surface electromyography.” Engineering in Medicine and Biology Society (EMBC), 2012 Annual International Conference of the IEEE. IEEE, 2012.

that can be downloaded for free at this link .pdf.

Preprocessing code can be found in the Materials.zip folder in src_1.0>scripts where there is synchro.m

Please note that a previous version of preprocessing source code can be found in Materials.zip folder in sincro.

Features Extraction

Features extraction code can be found in Materials.zip folder in ninapro_code.

To get an example of how this script works you can run demo_get_ninapro_data.m, while the function that is actually doing the work is get_ninapro_data.m.

You can choose among different features. However we recommend sticking to classical ones:

• Mean Absolute Value, type ‘mav’ in feature_type field
• ShortTime Fourier Transform, ‘stft’
• Variance, ‘var’
• Cepstral Coefficients, ‘cc’
• Waveform Length, ‘wl’

Once setted this paramethers, the code will do the hard work for you.

We strongly thank Ilja K. for features extracting code.

As for features extraction, we highly suggest reading Zecca and Micera work:

M. Zecca, S. Micera, M. C. Carrozza, and P. Dario, “Control of multifunctional prosthetic hands by processing the electromyographic signal,” Critical Reviews in Biomedical Engineering, vol. 30, no. 4-6, pp. 459–485, 2002.

If you feel the need for any further information, please feel free to contact us through the comment box below. We will be glad to help you through the path as soon as possible.

WHAT IS NINAPRO

This paper is about the latest NinaPro Database. We will briefly report its structure according to [1,2]. We strongly suggest to visit the NinaPro project website linked below.

The main aim of this project is to collect data among different subjects in order to improve the control of advanced hand prosthetics through surface electromyography. The NinaPro database shares its contribution in different fields and allows a large scale evaluation at reduced cost.

For the time being the set of possible movements with a prosthetic hand is restricted and it basically consists of simplified gestures such as opening and closing. Patients manage to control the hand after a long and exhausting learning process that causes loss of motivation and their consequent insucces.  As a result 30% to 50% of amputees do not use their prothesis regularly.

Database structure

The NinaPro project purpose was to build a database with a collection of sensor measurements taken on different subjects including intact  and amputated ones.

Further informations on the subjects are contained in each database explaining the range of age, weight and height.

Database 1:

27 intact subjects

Database 2:

40 intact subjects, consisting of 28 men and 12 women.

Database 3:

11 trans-radial aputated subjects.

Acquisition setup

The equipment used for data acquisition included:

 

1. Ciberglove II: this dataglove is composed by 22 strain gauges, and is used to gather the finger positions. In addition to the Cyberglove, a standard commercially available 2-axis inclinometer is fixed onto the subject’s wrist and used to collect the wrist orientation.
2. Surface EMG (sEMG): Muscular activity is gathered using ten sEMG electrodes. Because of the uniqueness of each amputation there was difference between positioning sEMG on intact subject and amputees ones. Thanks to pattern recognition techniques we can compensate little difference on sEMG placement and take advantage of muscle cross-talk, in order to use results of this research on different amputees subjects.

 

The experimental protocol:

The set of stimulus:

A set of 52 hand and wrist movements were selected basing on standard lits and previous similar experiments.

These movements were divided into four main classes:

1. 12 basic movements of the fingers (flexions and extensions);
2. 8 isometric, isotonic hand configurations (”hand postures”);
3. 9 basic movements of the wrist (adduction/abduction, flexion/extension and pronation/supination);
4. 23 grasping and functional movements — in this case, everyday objects are presented to the subject for grasping, in order to mimick a daily-life action.

The acquisition protocol:

During the data acquisition the subject had to replicate the movement he was seeing on a short movie in front of him.

The acquisition protocols on intact subjects and amputated ones were slightly different:

1. In the first case both the dataglove and the sEMG sensors were worn on the same limb.
2. In the second case the dataglove was worn on the intact limb and the sEMG sensors were worn on the amputees one. In this case the subject were asked to bilaterally perform the movement shown on the screen.

There were ten repetition of 5 seconds for each excercise, followed by three second of rest.

 

Data records

The database includes a set of .zip archives.

The archive contains three .mat (MATLAB) files, one for each excercise.

The variables in the files are:

• subject: subject number;
• excercise: excercise type;
• emg: sEMG measurements:
  • Columns 1-8: sEMG around the forearm;
  • Columns 9-10: sEMG on the Flexor and Extensor digitorum;
  • Columns 11-12 (if avaiable): sEMG on Biceps and Triceps muscles.
• acc: axis acceleration of the electrodes;
• glove: signals from the Cyberglove sensors;
• inclin: inclinometer values;
• stimulus: original label of the movement repeated;
• restimulus: a-posteriori refined label of the movement;
• repetition: stimulus repetition index;
• rerepetition: restimulus repetition index;
• force: force values;
• forcecal: force value boundaries (max e min limit);

 

For futher information on this topic, we strongly recommend to visit

[1] https://www.idiap.ch/project/ninapro

[2] http://ninapro.hevs.ch/node/16

from where we have taken all the information reported above.

Download .pdf