Beruflich Dokumente
Kultur Dokumente
I. INTRODUCTION
More indoor localization systems are being deployed as
wearable devices in entertainment, sports, and health. In
health care, for example, wearable inertial measurement units
(IMU), consisting of an accelerometer, a gyroscope, and a
magnetometer, is worn at the waist and fused with a Bluetooth
localization system to track the patients movements [1]-[3].
The key advantage of the IMU compared to existing tracking
systems is that they are highly portable, provide inertial data at
a high update rate, and do not suffer from a signal blockage.
However, estimating the position with the IMU alone leads to
the accumulation of the drift due to the instability of sensor
bias [4]. Some researchers have proposed the zero velocity
update (ZUPT) to correct the velocity error with a ground
contact, but the drift will still grow based on the total walking
distance [5].
The received signal strength indicator (RSSI) from a
low-power wireless technology such as Bluetooth Low
Energy (BLE) can be deployed to estimate a distance of a
receiver relative to a transmitter. With the multiple distance
measurements from the fixed anchor nodes (i.e. transmitters),
a trilateration can be used to estimate the absolute position of
the target node (i.e. receiver) [6]. However, the position
accuracy suffers from the non-line-of-sight (NLOS) and
multipath [7]. Therefore, the IMU and BLE measurements can
be fused to accurately estimate the drift-free position of the
The work is supported by the Natural Sciences and Engineering Research
Council of Canada (NSERC).
P. K. Yoon, S. Zihajehzadeh, and E. J. Park are with the School of
Mechatronic Systems Engineering, Simon Fraser University, 250-13450
102nd Avenue, Surrey, BC, Canada, V3T 0A3 (email: pky@sfu.ca,
szihajeh@sfu.ca, ed_park@sfu.ca).
B.-S. Kang is with the Department of Mechanical Engineering, Hannam
University, 70 Hannam-ro, Daejeon, 306-791, South Korea (email:
bskang@hnu.kr).
825
Accelerometer
Tilt KF
Gyr oscope
Yaw KF
Magnetome ter
ri [ xi x
ri ri
Position/
Velocity KF and
Smoother
IMU
BLE
z t Ct x t v t
yt
T
y t
tI 22
I 22
xt
1 2
t I 22
B t 1 2
tI
22
0 21
0 21
u t N a t NS R t S a t N g
Ct 1 I 22
I
A t 1 22
0 22
( xi x) 2 ( y i y ) 2 d i
ri , y
i 1
x t A t 1x t 1 B t 1u t 1 w t 1
x t xt
2
2
2
2
2
2
x x y y d d
n 1
n
n
n 1
n1 n
The residue between the estimated distance d i and the
distance to the location estimate x is calculated as following:
1n
ri , x
n i 1
ri
2( y1 y n )
2( x1 x n )
2( x n 1 x n ) 2( y n 1 y n )
ri
ri
xt
Pr B a log10 d
| yi y |]T
0 22
1 2
t I 23
w t 1 2
nA
tI
23
z t x BLE ,t
826
1
Q t t 2 I 22
2
1
tI 22 acc t 2 I 22
r 2
R t t,x
0
tI 22
rt2,y
Target
Node
Reference
Cameras
Anchor
Nodes
(a)
(b)
Figure 2. (a) Experimental lab setup with the target node, the anchor nodes,
and reference cameras and (b) Target node with the CC2540 BLE SoC with
the omnidirectional antenna, the MTx sensor, and an optical camera marker
I t1 (Pt1 ) 1
K t Pt ATt I t1
Pt Pt K t (Pt1 Pt 1 )K Tt
x t x t K t (x t 1 x t1 )
Figure 3. (a) Position estimate from the BLE trilateration, the proposed
algorithm, and the reference; and (b) residue on the Y- axis
827
(a)
(b)
Figure 6. Horizontal trajectory of (a) forward and (b) smoothed estimates
from t = 78 to 107.3s
(a)
(b)
Figure 4. (a) Absolute errors of trilateration estimations against the residues
on the X- and Y- axes and (b) Kalman gain of the position/velocity KF on the
standard KF and the proposed algorithm
Figure 5. (a) Forward and (b) smoothed position estimates from the standard
KF (KF), the proposed algorithm, and the reference on the Y- axis
TABLE I.
Filtering
Smoothing
KF
45.9
19.4
Proposed
36.7
18.1
V. CONCLUSION
In this paper, a novel three-step cascaded Kalman filter for
accurate estimation of the position trajectories with the IMU
and BLE trilateration measurement is proposed. Based on the
strong correlation between the trilateration residue and
trilateration error, the proposed algorithm uses the residue to
adaptively weight the trilateration estimate and the external
acceleration, thus the algorithm not requiring manual tuning
of the filter parameters. The experimental results have shown
that the proposed algorithm can accurately track the moving
sensor in the presence of the outliers, and the accuracy is
further improved by post-processing using the RTS smoother.
REFERENCES
[1]
828