Novel Velocity Update Applied for IMU-based

Wearable Device to Estimate The Vertical Distance

1st Tri Nhut Do 2nd U-Xuan Tan, Member, IEEE
Faculty of Engineering (FoE) Pillar of Engineering Product Development (EPD)
Van Lang University (VLU) Singapore University of Technology and Design (SUTD)
Ho Chi Minh City 700000, Vietnam Changi 487372, Singapore
trinhutdo@gmail.com uxuan tan@sutd.edu.sg

Abstract— The demand for indoor localization that does the coordination of the activities requires the tracking in real-
not rely on the presence of any external infrastructure had time of the operators for the rescuing and security
been increasing. In general, an indoor localization system was applications.
required to be precise, highly accurate and reliable. In this
paper, we presented and analyzed an indoor localization Two types of the infrastructure free localization system
wearable device that was capable of positioning people while are the strap-down and the step-and-heading. The strap-down
riding in an elevator. The inertial measurement unit (IMU) methods deal with displacement estimated by double integrat-
was utilized with an embedded system on the device. Current ing with reset in periods where the acceleration oscillates
approaches involving IMU mounted on a pedestrian’s body around zero level. The step-and-heading methods estimate
generally estimated the displacement on the ground only (in displacement by combining the step size that is calculated by
two dimensions). Thinking of a wearable device to estimate the a motion equation of an appropriate model with the yaw (user
vertical distance for elevator riding and with the fact that there heading) that estimated by quaternion-based filters [7]– [9].
are different levels of height for different buildings, a new The infrastructure free localization system utilizes sensors
algorithm was proposed to estimate distance in vertical such as IMU, force sensor, pressure sensor (barometer), ultra
direction when people riding in an elevator. The proposed wide band (UWB), etc. They can be worn on the user’s body
algorithm was based on the double integrating process from in the following ways: handheld [10], [11], mounted on foot
global acceleration with gravity removal in which the velocity
[12], [13], mounted at ankle [14], placed at the waist [15]–
and distance are updated in periods that the vertical
[17], chest and lower back [18]. The sensors and their
acceleration oscillates around Zero level. Experiments with a
wearable device which was designed based on the IMU model
positions are selected depending on the requirement of a
MPU9150, Arduino board and wireless Xbee took place for particular application.
riding in an elevator. Experimental results contained device’s There are a few approaches studied to estimate vertical
attitude, vertical distance and time stamp. They were recorded distance in indoor scenarios as such that of Diaz [19] who
online wirelessly via Xbee devices into an *.txt file. proposed an estimator for vertical displacement with IMU
Experiments in this work include riding up and down in placed in a pocket. The algorithm employed a linear equation
an elevator. They were repeated to collect data for
as the relationship between the pitch angle and the height of
evaluation by root mean square error (RMSE) computation
based on the ground-truth. The experimental results
stairs. The equation coefficients were obtained throughout
demonstrated RMSE of 0.77%, 0.88%, 1.66% riding in an many experiments recorded when the volunteer was climbing
elevator through one floor only, riding in an elevator through and descending stairs on the adjustable wooden staircases.
multiple floors while stopping at each floor, riding in an In this paper, we propose a novel method with no training
elevator through 40 floors, respectively. required to provide indoor vertical distance estimation. Build-
ing level heights are estimated based on double integration of
Keywords— Zero velocity intervals update, inertial naviga- vertical acceleration with reset at zero events in order to
tion, level height estimation eliminate accumulate error due to integration process. The
new method is applied for riding in an elevator. The proposed
I. INTRODUCTION position of Device is at people’s waist for convenient usage as
The localization system mentioned in this paper is used to shown in Fig. 1.
track the position of a user riding in an elevator without any
assistance needed from a pre-installed infrastructure on site
such as fiducial markers on the floor [1], infrared LED
landmarks on the panel [2], wireless communication devices
like radio frequency [3] and wireless local area network [4]–
[6]. Indoor localization is highly demanded for various
industries such as defense for operational planning or strategy
coordination of soldiers; social life such as public safety and
elderly care; and rescue applications like safety of firemen.
Indoor environments of such scenarios are more suitable for
the infrastructureless system employed due to no requirement
for on site hardware setup in advance, calibration and training.
Specifically, the rescuing and security applications are
often in unknown areas which are not suitable for
infrastructure- based system due to the fact that an external
infrastructure such as RFID, wireless network, infrared
sensors, landmark for vision cannot be installed. Moreover, Fig. 1. IMU is placed on the lower torso for greater convenience.

978-1-7281-3939-5/19/$31.00 ©2019 IEEE

Authorized licensed use limited to: University of Vermont Libraries. Downloaded on July 27,2020 at 02:17:37 UTC from IEEE Xplore. Restrictions apply.
 The rest of the paper is organized as follows. The proposed remains during zero intervals. The velocity update depicted
system and algorithm to estimate vertical distance (Level in the middle plot of Fig. 3 is formated in following equation:
Height in Building) are introduced in section II. In section III,
the experiment setting is described and the experimental  vi = vi−1 
results of a range of performed experiments are shown.
Section IV concludes the paper.
II. FLOOR HEIGHT ESTIMATION METHOD
Before any computation and estimation, all the coordinate
systems or frames must be assigned and two frames are
assigned as illustrated in Fig. 1. The subscript notations of W
and B indicate for World and Body respectively. The East-
North-Up (ENU) coordinate system is used as the world refer-
ence frame (XWYWZW) while the body coordinate system is
assumed to rigidly attach to the IMU and the IMU is mounted
at the people’s waist.
A. Indoor Inertial Navigation System
A quaternion-based Kalman Filter in indirect form [9],
[18] is employed to estimate 3D attitude by fusing the
information from the gyroscope, the accelerometer and the
magnetometer as depicted in the left part of Fig. 2. It consists
of the heading ( ˆ ), the pitch angle ( ˆ ) and the roll angle Fig. 3. The vertical integrating process when riding in an elevator up one
floor.
( ̂ ).
Fig. 3 depicts the integrating process in which the top
plot is the world frame vertical acceleration with gravity
removal discrete data when riding in an elevator up one
floor; the middle plot is integrated velocity and the bottom is
vertical integrated distance. The red dot value ’non-zero’ in
the top plot indicates the vertical acceleration zero intervals
where velocity integration is reset.
III. EXPERIMENTS
Fig. 2. Overview of the indoor Inertial Navigation System.
The capability of the proposed method is tested through
The employed attitude estimation algorithm has two ad- experiments conducted by four persons in our research group.
vantages as follows: 1) reducing computation cost due to In this section, a brief description of the experiment setup is
dealing with orientation errors instead of dealing directly first given, followed by a description of the experimental
with orientation which results in the state dimension being scenarios and the results obtained.
smaller and its response faster, and 2) two stages update A. Experimental Setup
using acceleration and magnetic strength by exploiting two The hardware used to test the proposed method is
measurements of the difference between accelerometer illustrated in Fig. 4, which consists of two parts: 1) the mobile
reading vs the gravity vector and the difference between part (attached to the pedestrian) shown in the top portion and
magnetometer reading vs the magnetic strength vector. 2) the base part shown in the bottom portion. The mobile part
B. Velocity Update consists of an Xbee wireless transmitter, an MPU-9150 IMU
The estimation for vertical distance when riding in an sensor board and an Arduino Fio microcontroller
elevator is obtained by double integrating the vertical ac- Atmega328P board, whereas the base part has an Xbee
celeration with gravity removal in world frame. In order wireless receiver connected to the laptop by virtual serial port
to remove accumulate error due to the integrating process, via USB port. The transmitter communicates with the Fio
velocity integration is reset during the defined zero velocity board using I2C communication protocol. The serial baud rate
intervals as in [1]. In this paper, an another new method to is set at 115200bps. The proposed algorithm is implemented
reset the velocity integration is proposed. It is reset during zero in the mobile part with the data acquired every 25 ms and
intervals of vertical acceleration instead of velocity passed through a complimentary filter with 225 ms time
integration. constant.

a) The Zero Detector: The mentioned zero intervals are

defined by a bound of two thresholds +0.05[m/s2] and
−0.05[m/s2 ]. These two thresholds are selected based on the
performance of accelerometer we utilized for doing
experiments, specially its recorded data during non- moving
status. All discrete vertical acceleration which are in the range
[−0.05; +0.05] are classified into zero intervals.
b) The Velocity Update: The velocity integrating Fig. 4. Diagram blocks of hardware for experiments, with the mobile part
process is temporarily stopped and its integrated value shown in the top and the base part shown in the bottom.

Authorized licensed use limited to: University of Vermont Libraries. Downloaded on July 27,2020 at 02:17:37 UTC from IEEE Xplore. Restrictions apply.
B. Experimental Results and Evaluation The last experimental scenario were taken place in another
Four members in the research group were required to place building named Block 90, where there are 20 staircases with
the device, which is shown in Fig. 5(a), at waist as depicted in 13.5cm height between two staircases. This produces the
Fig. 5(b) and to ride in an elevator up four times and down ground-truth of floor height equal to 2.7 meters. The vertical
four times. traveled journey for the last scenario is 108 meters which was
produced from 40 floors by 2.7 meters per floor. Therefore,
the root mean square error based on the ground-truth 108
meters is 1.7909 meters or 1.66%.
It is noted that the first two experiment scenarios (stopping
at each floor) result the closer root mean square error to each
other while the last experiment scenario results the higher one
(a) Device (b) Placed at waist than them due to longer integrating process of acceleration on
traveling through 40 floors.
Fig. 5. The device for vertical distance estimate.
IV. CONCLUSION
Three experimental scenarios were designed such as riding
This paper has proposed a novel method to update velocity
in an elevator through one floor only, riding in an elevator
during integrating process. A simple wearable device with
through multiple floors with one floor stops, riding in an
embedded algorithm is used for riding in an elevator to
elevator through 40 floors as shown in Fig. 3, Fig. 6, and Fig.
estimate floor height. The integrating process is updated
7, respectively.
during zero intervals of global vertical acceleration with
In order to evaluate the accuracy of vertical estimate, the gravity removal. We have created new method to update
staircase height was measured by Fluke 411D Laser Distance velocity and embedded it into a device to estimate vertical
Meter and the number of staircases between two floors were distance. The most convenient location of IMU for user is at
counted. waist and is provided in this paper as well. A range of
experiments were conducted by four subjects. Each subject
was tasked to perform the test four times. The RMSE is low
to 0.77% that proved the accuracy of the proposed method. In
this paper, all experiments are only focusing on riding in an
elevator. The distance estimations when people acts in other
motion modes such as standing, walking on an escalator will
be researched in future work.
ACKNOWLEDGMENT
This project is supported by Singapore University of Tech-
nology and Design (SUTD) Temasek Lab, Pillar of Engineer-
ing Product Development (EPD), Changi 487372, Singapore.
REFERENCES
Fig. 6. The vertical estimates are shown when riding in an elevator through [1] T. N. Do, and Y. S. Suh, “Gait Analysis Using Floor Markers and
multiple floors. Inertial Sensors,” Sensors, vol. 12, no. 2, pp. 1594–1611, 2012.
[2] T. N. Do, and Y. S. Suh, “Gait Analysis System Using Infrared LED
Landmarks,” Journal of Institute of Control, Robotics and Systems,
vol. 17, no. 7, pp. 641–646, 2011.
[3] P. Bahl, and V. Padmanabhan, “Radar: An in-building RF-based user
location and tracking system,” IEEE International Conference on Com-
puter Communications, pp. 775–784, 2000.
[4] H. Liu, H. Darabi, P. Banerjee, and J. Liu, “Survey of wireless indoor
positioning techniques and systems,” IEEE Transactions on Systems,
Man, and Cybernetics, Part C: Applications and Reviews, vol. 37, no.
6, pp. 1067–1080, 2007.
[5] K. El-Kafrawy, M. Youssef, A. El-Keyi, and A. Naguib, Propagation
modeling for accurate indoor wlan rss-based localization, IEEE 72nd
Vehicular Technology Conference Fall, pp. 1–5, Ottawa, Canada,
2010.
[6] L.-H. Chen, E. H.-K. Wu, M.-H. Jin, and G.-H. Chen, “Intelligent
Fig. 7. The vertical estimates are shown when riding in an elevator through fusion of Wi-Fi and inertial sensor-based positioning systems for
40 floors. indoor pedestrian navigation,” IEEE Sensors Journal, vol. 40, no. 11,
pp. 4034–4042, 2014.
The first two scenarios were taken place in a building [7] X. Yun, and E. R. Bachman, “Design, Implementation, and
in Singapore University of Technology and Design (SUTD), Experimental Results of a Quaternion-Based Kalman Filter for Human
where there are 30 staircases with 14.5cm height between two Body Motion Tracking,” IEEE Transactions on Robotics, vol. 22, no.
6, pp. 1216–1227, 2006.
floors. This produces the ground-truth of floor height equal to
[8] X. Yun, E. R. Bachman, and R. B. McGhee, “A Simplified Quaternion-
4.35 meters. The root mean square error (RMSE) is 0.77% and Based Algorithm for Orientation Estimation from Earth Gravity and
0.88% for one floor height experiment and for multiple floor Magnetic Field Measurements,” IEEE Transactions on Instrumentation
heights experiment, respectively. and Measurement, vol. 57, no. 3, pp. 638–650, 2008.

Authorized licensed use limited to: University of Vermont Libraries. Downloaded on July 27,2020 at 02:17:37 UTC from IEEE Xplore. Restrictions apply.
[9] Y. S. Suh, “Orientation estimation using a quaternion-based indirect Conference on Robotics and Biomimetics, pp. 2115–2120, China, Dec.
Kalman filter with adaptive estimation of external acceleration,” IEEE 2015.
Transactions on Instrumentation and Measurement, vol. 59, no. 12, [15] A. Kose, A. Cereatti, and U. D. Croce, “Estimation of traversed
pp. 3296–3305, 2010. distance in level walking using a single Inertial Measurement Unit
[10] V. Renaudin, M. Susi, G. Lachapelle, “Step Length Estimation using attached to the waist,” IEEE International Conference in Medicine and
Handheld Inertial Sensors,” Sensors, vol. 12, no. 7, pp. 8507–8525, Biology Society, pp. 1125–1128, 2011.
2012. [16] J. C. Alvarez, D. Alvarez, A. Lope?z, and R. C. Gonza?lez, “Pedestrian
[11] H. Zhang, W. Yuan, Q. Shen, T. Li, and H. Chang, “A Handheld Navigation Based on a Waist-Worn Inertial Sensor,” Sensors, vol. 12,
Inertial Pedestrain Navigation System with Accurate Step Modes and no. 8, pp. 10536–10549, 2012.
Device Poses Recognition,” IEEE Sensors Journal, vol. 15, no. 3, pp. [17] F. Inderst, F. P. Santoni, “3D pedestrian dead reckoning and activity
1421–1429, 2015. classification using waist-mounted inertial measurement unit,” IEEE
[12] H. M. Schepers, E. H. van Asseldonk, C. T. Baten, and P. H. Veltink, International Conference on Indoor Positioning and Indoor Navigation
“Ambulatory estimation of foot placement during walking using (IPIN), pp. 1–9, 2015.
inertial sensors,” Journal of Biomechanics, vol. 43, no. 16, pp. 3138– [18] T. N. Do, R. Liu, C. Yuen, and U-X. Tan, “Personal Dead Reckoning
3143, 2010. using IMU mounted on upper Torso and inverted Pendulum Model,”
[13] E. Bachmann, J. Calusdian, E. Hodgson and X. Yun, “In situ heading IEEE Sensors Journal, vol. 16, no. 21, pp. 7600–7608, 2016.
drift correction for human position tracking using foot-mounted iner- [19] E. M. Diaz, “Inertial Pocket Navigation System: Unaided 3D Position-
tial/magnetic sensors,” IEEE International Conference on Robotics and ing,” Sensors, vol. 15, no. 4, pp. 9156–9178, 2015.
Automation (ICRA), pp.5425–5430, Saint Paul, MN, May. 2012.
[14] T. N. Do, R. Liu, C. Yuen, and U-X. Tan, “Design of an infrastruc-
tureless in-door localization device using an IMU,” IEEE International

Authorized licensed use limited to: University of Vermont Libraries. Downloaded on July 27,2020 at 02:17:37 UTC from IEEE Xplore. Restrictions apply.