Page 43 - Fister jr., Iztok, and Andrej Brodnik (eds.). StuCoSReC. Proceedings of the 2018 5th Student Computer Science Research Conference. Koper: University of Primorska Press, 2018
P. 43
ards an enhanced inertial sensor-based step length
estimation model
Melanija Vezocˇnik Matjaž Branko Juricˇ
University of Ljubljana, Faculty of Computer and University of Ljubljana, Faculty of Computer and
Information Science Information Science
Ljubljana, Slovenia Ljubljana, Slovenia
mv6082@student.uni-lj.si matjaz.juric@fri.uni-lj.si
ABSTRACT models require acceleration, angle-based models require the
opening angle of the leg and the input of multiparameter
Inertial sensors found in Internet-of-Things devices such as models combines two or all three of the previously listed
smartphones and smartwatches can be used to track user’s parameters.
activity as well as step length. In particular, rapidly grow-
ing interest in pedestrian dead reckoning – indoor position- In this work we present our research towards an enhanced
ing approach that calculates current position as the sum of inertial sensor-based step length estimation model that in-
previous position and vector with step length and heading creases the accuracy of estimated step length. Section 2
information – prompts to develop refined step length esti- overviews step length estimation models. Section 3 describes
mation models to overcome inaccuracy in determining step the derivation and Section 4 the evaluation of the proposed
length. In this work we present our research towards an en- model. Section 5 presents results, whereas Section 6 dis-
hanced inertial sensor-based step length estimation model. cusses them. Finally, Section 7 concludes this work present-
For this purpose, we extend an existing step-frequency-based ing future research directions.
model with acceleration and make it less user-specific. We
evaluated the performance of the proposed model for one 2. OVERVIEW OF STEP LENGTH ESTIMA-
sensor position and different walking speeds and obtained TION MODELS
very promising results comparable to related models. We
will further improve the accuracy of the proposed model Step length is determined in several phases. In order to
and modify it, so it will require less time for tuning, and acquire inertial sensor measurements, a sensor has to be
thoroughly address the pre-processing of the sensor data. set-up on user’s body, most commonly in hand, pocket or
We will also test the proposed model for different sensor centre of body mass. Acquired data – usually accelerome-
positions and pedestrian-dead-reckoning-based indoor posi- ter measurements – have to be pre-processed to eliminate
tioning. errors. Steps are detected before step length is determined
using a step length estimation model that often has to be
Keywords tuned. Next, we briefly present the categories of the models
(step-frequency-based, acceleration-based, angle-based and
inertial sensors, pedestrian dead reckoning, step length esti- multiparameter).
mation
2.1 Step-frequency-based models
1. INTRODUCTION
Step-frequency-based models require step frequency as an
Over the past few years, user engagement enabling Internet- input. They commonly exploit linear relationship between
of-Things (IoT) technologies are emerging. In particular, in- step length and step frequency and have to be tuned prior to
ertial sensors can be utilized to track user’s activity as well as utilization. The following two models include user’s height.
step length since they can be found in smartphones, smart- Renaudin et al. [6] proposed a model based on linear rela-
watches and other IoT devices. They also exhibit advan- tionship between step length and step frequency, whereas
tages such as easy accessibility, inexpensiveness and small- Tian et al. [2] exploited the link between square root of step
size. Moreover, estimating step length is also applicable frequency and step length. On the contrary, Zhang et al. [1]
in diverse areas, especially in indoor positioning approach included user’s leg length in their model, but based it on the
called pedestrian dead reckoning [1, 2, 3, 4] that determines linear relationship between step length and step frequency,
the current position by adding vector with step length and similarly as the model proposed by Renaudin et al. [6].
heading information to the previous position.
2.2 Acceleration-based models
Throughout this work we address inertial sensor-based step
length estimation models. According to [5], they can be Acceleration-based models require acceleration as an input.
classified into four categories based on the parameters they They often exploit the link between step length and accel-
require as an input. These parameters are obtained from eration properties such as minimum and maximum accel-
inertial sensor measurements. The categories are: step- eration values within the step. Weinberg [7] proposed such
frequency-based [1, 2, 6], acceleration-based [3, 7], angle- model. It calculated step length as the product of a tuneable
based [8] and multiparameter [4, 9]. Step-frequency-based constant and fourth principal root of the difference between
models require step frequency as an input, acceleration-based
StuCoSReC Proceedings of the 2018 5th Student Computer Science Research Conference DOI: https://doi.org/10.26493/978-961-7055-26-9.43-46 43
Ljubljana, Slovenia, 9 October
estimation model
Melanija Vezocˇnik Matjaž Branko Juricˇ
University of Ljubljana, Faculty of Computer and University of Ljubljana, Faculty of Computer and
Information Science Information Science
Ljubljana, Slovenia Ljubljana, Slovenia
mv6082@student.uni-lj.si matjaz.juric@fri.uni-lj.si
ABSTRACT models require acceleration, angle-based models require the
opening angle of the leg and the input of multiparameter
Inertial sensors found in Internet-of-Things devices such as models combines two or all three of the previously listed
smartphones and smartwatches can be used to track user’s parameters.
activity as well as step length. In particular, rapidly grow-
ing interest in pedestrian dead reckoning – indoor position- In this work we present our research towards an enhanced
ing approach that calculates current position as the sum of inertial sensor-based step length estimation model that in-
previous position and vector with step length and heading creases the accuracy of estimated step length. Section 2
information – prompts to develop refined step length esti- overviews step length estimation models. Section 3 describes
mation models to overcome inaccuracy in determining step the derivation and Section 4 the evaluation of the proposed
length. In this work we present our research towards an en- model. Section 5 presents results, whereas Section 6 dis-
hanced inertial sensor-based step length estimation model. cusses them. Finally, Section 7 concludes this work present-
For this purpose, we extend an existing step-frequency-based ing future research directions.
model with acceleration and make it less user-specific. We
evaluated the performance of the proposed model for one 2. OVERVIEW OF STEP LENGTH ESTIMA-
sensor position and different walking speeds and obtained TION MODELS
very promising results comparable to related models. We
will further improve the accuracy of the proposed model Step length is determined in several phases. In order to
and modify it, so it will require less time for tuning, and acquire inertial sensor measurements, a sensor has to be
thoroughly address the pre-processing of the sensor data. set-up on user’s body, most commonly in hand, pocket or
We will also test the proposed model for different sensor centre of body mass. Acquired data – usually accelerome-
positions and pedestrian-dead-reckoning-based indoor posi- ter measurements – have to be pre-processed to eliminate
tioning. errors. Steps are detected before step length is determined
using a step length estimation model that often has to be
Keywords tuned. Next, we briefly present the categories of the models
(step-frequency-based, acceleration-based, angle-based and
inertial sensors, pedestrian dead reckoning, step length esti- multiparameter).
mation
2.1 Step-frequency-based models
1. INTRODUCTION
Step-frequency-based models require step frequency as an
Over the past few years, user engagement enabling Internet- input. They commonly exploit linear relationship between
of-Things (IoT) technologies are emerging. In particular, in- step length and step frequency and have to be tuned prior to
ertial sensors can be utilized to track user’s activity as well as utilization. The following two models include user’s height.
step length since they can be found in smartphones, smart- Renaudin et al. [6] proposed a model based on linear rela-
watches and other IoT devices. They also exhibit advan- tionship between step length and step frequency, whereas
tages such as easy accessibility, inexpensiveness and small- Tian et al. [2] exploited the link between square root of step
size. Moreover, estimating step length is also applicable frequency and step length. On the contrary, Zhang et al. [1]
in diverse areas, especially in indoor positioning approach included user’s leg length in their model, but based it on the
called pedestrian dead reckoning [1, 2, 3, 4] that determines linear relationship between step length and step frequency,
the current position by adding vector with step length and similarly as the model proposed by Renaudin et al. [6].
heading information to the previous position.
2.2 Acceleration-based models
Throughout this work we address inertial sensor-based step
length estimation models. According to [5], they can be Acceleration-based models require acceleration as an input.
classified into four categories based on the parameters they They often exploit the link between step length and accel-
require as an input. These parameters are obtained from eration properties such as minimum and maximum accel-
inertial sensor measurements. The categories are: step- eration values within the step. Weinberg [7] proposed such
frequency-based [1, 2, 6], acceleration-based [3, 7], angle- model. It calculated step length as the product of a tuneable
based [8] and multiparameter [4, 9]. Step-frequency-based constant and fourth principal root of the difference between
models require step frequency as an input, acceleration-based
StuCoSReC Proceedings of the 2018 5th Student Computer Science Research Conference DOI: https://doi.org/10.26493/978-961-7055-26-9.43-46 43
Ljubljana, Slovenia, 9 October
