Page 146 - Control Theory in Biomedical Engineering

Modeling simple and complex handwriting based on EMG signals 131

2 History of handwriting modeling

In 1962, Van Der Gon proposed a second-order model allowing to assim-
ilate the hand as a mass moving on the writing surface (Van Der Gon et al.,
1962). An electronic version of that model was then proposed by Mac
Donald that presented the handwriting process as a dynamic mass that moves
in a viscous environment (Mac Donald, 1964). The movement of this
mass was described by a linear differential equation of the second order.
In 1975, Yasuhara showed that the vertical movements of the pen tip are
generated by the flexion movements of the forearm muscles, while horizon-
tal displacements are generated by the abduction-adduction movements of
these muscles. Then, he integrated the effect of the friction force between
the tip of the pen and the writing surface, (Yasuhara, 1975) from which he
identified and composed a fast writing system (Yasuhara, 1983). From this
model, Iguider and Yasuhara developed, in 1995, two approaches, one for
the extraction of pulsations of control (Iguider and Yasuhara, 1995) and the
other, in 1996, for the recognition of cursive Arabic script (Iguider and
Yasuhara, 1996). As the muscles of the forearm intervening in the hand-
writing act are directly located under the skin, surface electrodes were
used to record the generated EMG signals. Based on this possibility, hand-
writing experimentation was realized at Hiroshima University to measure
two forearm muscles’ activities concurrently with the coordinates of
some Arabic letters and geometric forms (Manabu et al., 2003). Based on
this experimental approach and dataset, different mathematical approaches
have been proposed to characterize the handwriting process from EMG
signals.
Manabu et al. (2003) proposed a third-order model to generate writing
from two EMG signals. Linderman et al. (2009) proposed a handwriting
model from four EMG signals to produce handwritten numbers from zero
to nine. Using the same experimental measurements, Okorokova et al.
(2015) developed a hybrid model, based on the KF and the model proposed
by Mac Donald (1964). Moreover, Zhang and Kamavuako (2019) devel-
oped, based on their own recording of three EMG channels, a neural net-
work model based on an experimental approach allowing recording four
geometric forms drawing on an (x, y) plane. These diverse writing charac-
terization models have contributed to obtaining more or less satisfactory
results requiring a parametric adjustment, related to each change of
inputs/outputs to be modeled, limiting the efficiency of these structures,
mainly due to the subjective nature of the EMG signals. In order to mitigate