Page 161 - Control Theory in Biomedical Engineering
P. 161
Modeling simple and complex handwriting based on EMG signals 145
• Lower observer e X:
8
>
>
<
e Xk +1ð Þ ¼ A i XkðÞ + B i UkðÞ + K i Y e Y
e
(11)
>
>
:
e YkðÞ ¼ C i XkðÞ
e
8
X e kðÞ ¼ αXkðÞ + βXkðÞ
e
e
<
(12)
:
Y e kðÞ ¼ αYkðÞ + βYkðÞ
e
e
with:
½
α 0, 1
β 0, 1,α + β ¼ 1
½
The estimated coordinates are then calculated using the relationships
from Eqs. (3)–(7).
For Figs. 9 and 10, the dotted blue line (dotted gray line in print version)
shows the experimental data and the solid blue line (solid gray line in print
version) is relative to the outputs of the model.
6Discussion
In this chapter, we presented different models that have been proposed to
characterize handwriting from EMG signals. The first model is based on
the KF to reproduce numbers zero to nine from eight EMG channels. This
complex model requires the estimation of 48 parameters and fixing specific
hypotheses that may not be respected in real-time development. Besides, this
approach is constituted by the fusion of two models, which increases the
computational time. The results of this approach show a significant error
between estimated and desired data. The second model is based on the neu-
ral network approach to estimate some geometric shapes from three EMG
signals. The results of this approach are acceptable, however, it shows that
the model needs refinement. This model is characterized by a moderate
computational time during training only because of the single hidden layer.
In order to model handwriting of Arabic letters and geometric forms from
only two EMG signals recorded from the forearm muscles, Manabu et al.
(2003) proposed a dynamic third-order model to characterize pen-tip move-
ments on an (x, y) plane. However, the model is limited in terms of accuracy
and computational time.
Unlike these handwriting models, the advantage of the modeling
approach based on interval observer is to reduce the number of inputs
