Multiparameter Sleep Monitoring Using a Depth Camera - Biomedical Engineering Systems and Technologies - page 298

system could not recognize
hand position, and then we
ping by hand.

e it as a head. One possible solution method is to detect
e can estimate the head position while the head is overl

the
lap-

Fig. 11. Special Sleep Postures

In addition, there are so
discuss. First, the algorithm
detect multiple heads and t
used to pair the head and
detection problem while the
detect in the prone position
in the supine position or pr
head and torso detection in

ome sleep conditions or sleep positions that we did
m of head and torso detection we proposed can be applied
torsos. Moreover, a shortest distance pairing procedur
d torso of specified sleeper. However, we still have
e users are overlapping. Second, the head and torso can
. Howerer, this system can not recogize whether the use
one position now. Fig. 12 shows the detection result of
the prone position and multiple users.

not
d to
e is
the
n be
er is
f the

Conditions. Left: multiple sleepers. Right: prone position Fig. 12. Other Sleep C

5.2 Breath Measureme ent

From the experimental resu
which was similar to the con
accuracy of breathing is hig
lying position. We observed
the supine position. Besid
amplitude of the breathing s
age amplitude of the breathi
all, less signal noises and m
accuracy while sleeping in t
racy of breathing while sle
sleeping with no quilt. We s
reduces the wrinkles of the t
breathing signal, but the sy
breathing exercise or other e
results in experiment I, we f
moving or turning around i

ults of breath measurement, we observed some phenome
nditions of head and torso detection. First, the measurem
gher while sleeping in the supine position than in the si
d that there are more noises in the side-lying position than
des, according to our measurement results, the aver
signals is 0.8 cm in the supine position. However, the av
ing signals is about 0.5 cm in the side-lying position. Ov
more breathing amplitude would increase the measurem
the supine position. Second, the overall measurement ac
eping with a thin quilt was better than the accuracy wh
speculate that it might because the reason that the thin q
torso surface. Third, the movement of the torso is seen as
ystem cannot identify whether the movement is caused
exercises. According to the observations of the measurem
found that there were more detection errors while the use
in bed. In experiment I, the participants were asked to t

non
ment
ide-
n in
rage
ver-
ver-
ment
ccu-
hile
quilt
the
d by
ment
er is
turn