Vous êtes sur la page 1sur 2

On the Detection of Independent Finger Movements

through Two Forearm Myoelectrodes


P. Steele, J. Brooks, N. Gomes, Y. Sun, E. Chabot
University of Rhode Island
45 Upper College Road
Kingston, RI 02881

Abstract This study analyses forearm muscle contractions that generate one EMG measurement with a third electrode being a
occur during movements of individual fingers. An amplitude ground reference. The electrodes are placed on the belly of the
threshold is implemented on the multiplied backwards difference flexor carpi radialis, the belly of the palmaris longus, and,
(MOBD) of the signal to determine when a muscle contraction lastly, to the area slightly proximal of the olecranon, serving
occurs, and is followed by a windowed variance threshold of the
as a driven ground. Signal conditioning is performed by
MOBD to determine which finger contracted. The purpose is to
create a low-cost, computationally efficient method to determine differentially amplifying the electrodes by a gain of 50. Prior
specific finger contractions from the forearm. This method was to digitization, the signal is high pass filter at 7 Hz and notch
able to differentiate the ring finger from the other fingers, which filtered at 60 Hz. Data is sampled at a rate of 250 Hz.
in comparison had a variance approximately ten times higher.
B. Finger Discrimination Algorithm
I. INTRODUCTION With a digitized signal, an algorithm is proposed for the
Approximately 250,000 Americans have a spinal cord detection of muscle activation and discrimination of
injury, with 12,000 new injuries occurring each year. contracting finger. First, a multiplication of backward
Therefore, there is a growing need for assisting these patients, differences (MOBD) algorithm is applied by differencing
specifically those with paraplegia. A common method for sampled data as follows:
increasing the quality of life of individuals with paraplegia is
the power wheelchair. However, a joystick, which is one of [] = [] [ 1]
the few readily available mechanisms of control, may not be
feasible for individuals with limited arm mobility due to a where u is the input samples and n is the sample number. If
reduced range of motion. Many proposed electromyogram u[n] and u[n-1] have differing signs, the resultant value of the
(EMG) based control methods distinguish between different difference, x[n], is defined as 0. Next, backward differences,
contractions including activation site, pattern, and amplitude x[n], are multiplied based on a MOBD order of N=2
leverage complex algorithms[1] and multiple electrodes[2],
1
the latter two of which make these methods more costly and
intrusive on the user. A new method is presented that utilizes [] = [ ]
one differential measurement combined with a multiplication =0
of backward differences (MOBD) algorithm based upon work
by Suppappola [3] to simplify hardware and software design. Any threshold exceedance of y[n] is defined as an
activation, where the threshold is heuristically set at 0.01V.
Due to the random signal fluctuations, an activation threshold
crossing is maintained for R samples subsequent.
Experimentally, this was defined as 100 samples (or 2/5ths of
a second).

Further extending the process, the variance of the MOBD is


calculated over a moving window of size M as
1
1
[] = |[ + ] |2

=0
where is defined as the mean of the windowed samples.
Figure 1. Flow Diagram of Signal
Subsequently, a threshold is applied to the variance to classify
II. METHODS
different types of finger based activations over a window. The
A. Instrumentation Overview window size is selected based up an acceptable
For the proposed technique, two electrodes are used to misclassification rate and desired threshold level. A larger
window can reduce the misclassification rate and allow for an
978-1-4799-8360-5/15/$31.00 2015 IEEE. increased variance threshold. Choosing a large window size
also comes at the expense of a decision delay, which reduces Computationally complex algorithms, such as Multiclass
responsiveness of the interface. Support Vector Machine (SVM) based algorithms, have been
shown to work with high classification accuracy for each
III. .RESULTS
individual finger activation and robustness to subject variation
Sample data was recorded from each finger by contracting [4], but are limited in potential applications due to system
and releasing the finger, alternating approximately every two complexity and configuration demands. With advances in
seconds, in a ten second sample period. In figure 2, an technology, these methods may demonstrate more feasibility
example data set is presented of the smallest finger with the in low-power or long endurance applications.
resultant MOBD value and activation detection result. In
figure 3, the fraction of points above the variance threshold is Further work is anticipated to refine this method by changes
illustrated. to hardware and algorithm design. While the technique
described could not reliably isolate all fingers, an expansion of
2
Raw Signal
the approach could also be explored with additional electrodes
MOBD and modified electrode placement. Even with the addition of
1.5 On/off Detection
electrodes, the system would likely fit in a low power
microcontroller environment. Additionally, an algorithm for
1 automatic threshold adjustment based upon the individual
should be explored.
Amplitude

0.5

0.8
Index
0
0.7 Middle
Ring
Pinky
-0.5 0.6

Fraction Above Threshold 0.5


-1
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0.4
Sample
Figure 2. This graph represents the original data of the ring finger with 0.3
the bias removed (red), the MOBD performed on the data (cyan), and the
detection of a contraction by the variance algorithm (black). 0.2

IV. CONCLUSIONS 0.1


An MOBD algorithm has been proposed which is
computationally efficient and allows for discrimination of 0
0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.01
finger activation. From preliminary data captured, the results Threshold
show that muscle activations associated with different finger
movements have distinct statistics. The different distributions Figure 3. This graph represents the fraction of points above a threshold.
This illustrates the ability to discriminate between the ring finger and the
of variance are used to isolate the ring finger from the index, other classes. The error bars represent the minimum and maximum
middle, and smallest fingers. With a threshold of 0.003 on the bounds of the individual trials.
variance, the ring finger data reflects greater than 10% of the
active points above the threshold. Correspondingly, all other ACKNOWLEDGEMENT
fingers dont have any data points above the threshold This work was fully supported by the University Of Rhode
resulting in no false alarms. Based upon the average rate of Island Office Of Research Development through the URI
points above the threshold, a decision window would need to Undergraduate Research Initiative Program.
be greater than 10 points. A larger window should be selected
REFERENCES
to reduce the associated risk of a misclassification trade-off
with the decision delay of the window length. [1] S. Han et al, Human-Machine Interface for wheelchair control with EMG
and its evaluation, Proceedings of the 25th International Conference of
This simple method can potentially be used in the future to the IEEE EMBS, Cancun, 2003.
control assistive technology devices through the use of EMG [2] X. Xu et al.,Robust Bio-Signal Control of an Intelligent Wheelchair,
with the mapping of each finger to a direction or movement. Journal of Robotics, 2013. Print.
For example, contracting the smallest finger could be used to [3] S. Suppappola and Ying Sun, 'Nonlinear transforms of ECG signals for
move forward, contracting the ring finger could be used to digital QRS detection: a quantitative analysis', IEEE Transactions on
move backwards, and once the middle and index fingers have Biomedical Engineering, vol. 41, no. 4, pp. 397-400, 1994.
been distinguished separately, contracting the middle finger [4] S. Maier and P. van der Smagt, 'Surface EMG suffices to classify the
could be used to move left, and index finger could be used to motion of each finger independently', in 9th International Conference on
Motion and Vibration Control, Munich, 2008.
move right.

Vous aimerez peut-être aussi