sensors

Article
Wireless Indoor Localization Using Convolutional
Neural Network and Gaussian Process Regression
Guolong Zhang 1 , Ping Wang 2 , Haibing Chen 1 and Lan Zhang 1, *
1 School of Automation and Electronic Engineering, University of Science and Technology Beijing,
Beijing 100083, China; [email protected] (G.Z.); [email protected] (H.C.)
2 Space Star Technology Co., Ltd., Beijing 100086, China; [email protected]
* Correspondence: [email protected]




Received: 8 April 2019; Accepted: 28 May 2019; Published: 31 May 2019


Abstract: This paper presents a localization model employing convolutional neural network (CNN)
and Gaussian process regression (GPR) based on Wi-Fi received signal strength indication (RSSI)
fingerprinting data. In the proposed scheme, the CNN model is trained by a training dataset. The
trained model adapts to complex scenes with multipath effects or many access points (APs). More
specifically, the pre-processing algorithm makes the RSSI vector which is formed by considerable RSSI
values from different APs readable by the CNN algorithm. The trained CNN model improves the
positioning performance by taking a series of RSSI vectors into account and extracting local features.
In this design, however, the performance is to be further improved by applying the GPR algorithm to
adjust the coordinates of target points and offset the over-fitting problem of CNN. After implementing
the hybrid model, the model is experimented with a public database that was collected from a library
of Jaume I University in Spain. The results show that the hybrid model has outperformed the model
using k-nearest neighbor (KNN) by 61.8%. While the CNN model improves the performance by
45.8%, the GPR algorithm further enhances the localization accuracy. In addition, the paper has also
experimented with the three kernel functions, all of which have been demonstrated to have positive
effects on GPR.

Keywords: fingerprinting localization; received signal strength indication; k-nearest neighbor;

convolutional neural network; Gaussian process regression; cumulative error distribution

1. Introduction
With the rapid growth of the Internet of Things market, indoor localization has long been a
question of great interest in a wide range of fields. There is an urgent need to address the precise
indoor localization problems caused by location-based services. Location-based services typically
include indoor navigation, shop finding, targeted advertising, transportation, users flow analysis
and other industrial fields [1–3]. For localization in an outdoor environment, a Global Navigation
Satellite System (GNSS) is an ideal method that meets people’s performance requirements. However,
the signals from GNSS have proven to be unreliable in an indoor environment. Therefore, we urgently
need positioning methods that can perform well in indoor environments [4,5].
Various methods applied to provide a localization service have different accuracies in different
kinds of indoor environments. Existing research recognizes the critical role played by received signal
strength indication (RSSI) based localization methods. Many RSSI based systems require periodic
calibrations and information fusion which decreases their performance [6]. The RSSI-based localization
method, whether ranging-based or not, is still the mainstream research method. For example, a
RSSI ranging-based algorithm utilizing the radio and search and the least squares algorithm can be
used for partial discharge source localization [7]. Besides, Wi-Fi fingerprinting localization is one

Sensors 2019, 19, 2508; doi:10.3390/s19112508 www.mdpi.com/journal/sensors

Sensors 2019, 19, 2508 2 of 18

of the methods based on RSSI in wireless sensor networks (WSNs) [8,9]. Compared to other indoor
localization methods, Wi-Fi fingerprinting localization technology has some advantages including
low hardware requirement and wide scope of application. At the same time, the technology needs to
cooperate with more advanced algorithms to ensure higher positioning precision.
Fingerprinting-based localization algorithms are usually either deterministic or probabilistic
algorithms. As for deterministic algorithms, previous studies have successfully manifested the crucial
role played by the k-nearest neighbor (KNN) algorithm and its variants including the weighted KNN
(WKNN) algorithm in a fingerprinting-based localization area [10]. In [11], Shin et al. found that the
enhanced WKNN reduces the error compared to KNN by adjusting the number of considered neighbors.
In [12], Fang et al. proposed an optimal WKNN algorithm composed of an adaptive Kalman filter and
a memetic algorithm. These advanced algorithms based on WKNN have improved the localization
performance compared to the KNN algorithm. Besides, previous studies have revealed that support
vector machines (SVM) are also an effective means to solve Wi-Fi fingerprinting localization. Abdou
et al. [13] presented the combining of SVM and cluster as a regression localization algorithm. The
algorithm first analyzed the data in the fingerprinting database by clustering and then employed SVM
to establish the intrinsic relationship between locations and fingerprints. However, when the dimension
of the RSSI vectors is large, the SVM algorithm will significantly increase in time complexity. As for
probabilistic algorithms, they commonly build a probability distribution model on the RSSI values in
the fingerprinting database firstly and then apply probabilistic algorithms to estimate the locations of
test points (TPs) [14–16]. In [17], Piotr et al. focused on the Kullback-Leibler divergence metric that
compared multivariate RSSI distributions to provide accurate location estimates. A Gaussian process
(GP) was presented as a likelihood model for RSSI values [18]. Many probabilistic algorithms, including
Gaussian processes, give confidence intervals for predicted values, however do not achieve ideal
localization accuracy. Therefore, the research of KNN-based localization algorithms is more common.
Prior studies have not been able to convincingly manifest that a proposed algorithm will achieve
an ideal performance in real-world scenarios. With a focus on reflecting the effect of the algorithm in a
complex environment, we chose a fingerprinting database collected in the library of Jaume I University
as the data source of this paper. Additionally, high-dimensional RSSI vector in this database can better
verify the performance of the algorithm.
In this paper, we propose a wireless positioning hybrid model using both convolutional neural
networks (CNN) and Gaussian process regression (GPR). Recently, CNN has been applied to related
research in indoor localization. Several studies, for instance [19] and [20], were carried out on
investigating the effectiveness of using CNN for visual localization. CNN is commonly employed for
image recognition because they could be thought of automatic feature extractors of images by using
adjacent pixel information to effectively subsample the image first by convolution and then adopting
a prediction layer at the end [21]. Similarly, we set the fingerprinting dataset as several “image”. A
number of RSSI vectors of a location at different times were chosen to form an “image”. Each RSSI
vector was treated as a multi-channel pixel in the “image”. Finally, each RP in the fingerprinting dataset
got its own “image” which will be utilized in the further algorithm. In the first phase of the model,
we used CNN to extract reliable features of “images” and then built internal representation between
“images” and locations of reference points (RPs) based on the Pytorch computational framework. In
the second phase, we evaluated the estimation error of the trained CNN model by validation points
(VPs), then trained the GPR algorithm by the error with their respective fingerprints of VPs. Finally,
we employed a trained GPR model to correctly estimate coordinates of CNN in order to improve the
location accuracy. GPR is a widely accepted algorithm for solving nonlinear regression problems
for its probabilistic benefits and ability to put high-dimensional data within reach. Combining the
advantages of CNN and GPR algorithms, our proposed hybrid model effectively mitigates multipath
effects, other noises and human body interference on Wi-Fi signals, improving positioning accuracy.
The main contribution of this paper is as follows:

• A fingerprinting-based localization hybrid model using both the CNN and GPR algorithm.
Sensors 2019, 19, 2508 3 of 18

• Performance evaluation of the proposed model and its comparison with the KNN algorithm.
• The comparison of positioning performance between the hybrid model based on different
kernel functions.

The remaining part of the paper proceeds as follows: Section 2 begins by introducing the Wi-Fi
fingerprinting localization technology analysis and then analyses the data gathered, focusing on signals
pre-processing. Section 3 is concerned with the proposed model including the structure and other
details about CNN and GPR models. In particular, Section 3.5 contextualizes the GPR model by
fingerprinting-based localization and highlights the key concept. Section 4 presents the experimental
results and discusses the significant findings. The final section gives a summary and critique of
the findings.

2. Wi-Fi Fingerprinting Localization

2.1. Technology Analysis

In recent years, there has been an increasing interest in algorithms based on fingerprinting
localization technology [22]. The reason for the above phenomenon is related to the characteristics of
fingerprinting localization. On the one hand, it has the advantages of being low-cost and having wide
applications in indoor scenarios. A general indoor scenario where enough intense Wi-Fi networks are
deployed without any other hardware could be an ideal place to apply the technology. On the other
hand, the disadvantages of the technology include high requirements for data support and algorithm
quality to ensure accuracy. Admittedly, there are two types of solutions including increasing collecting
density of RPs and improving the algorithm to find better matching function between locations of RPs
and their respective RSSI vectors [23]. From the aspect of practice, the former solution will increase
costs which come from setting more access points (APs) in advance, taking more time to collect the
fingerprinting data and processing more complex information. Consequently, paying more attention
to an approved algorithm is a more economical choice. In addition, a specific fingerprinting database
is required to solve the localization problem of a particular scene. Once the localization environment
changes, the fingerprinting database needs to be re-acquired.
Other than that, there are some influence factors of Wi-Fi fingerprinting localization technology.

• The position and movement of the human body in an indoor environment can trigger signal
fluctuations [24]. This is even more crucial in public areas where there is a large flow of people [25].
As is exhibited in Figure 1, the RSSI values from three APs fluctuate over a period of time. It
manifests that the RSSI value is not very stable.
Sensors 2019, 19, 2508 4 of 21

1. The1.variation
FigureFigure of received
The variation signal
of received strength
signal strengthindication (RSSI)values
indication (RSSI) values from
from three
three access
access points
points (APs).(APs).

As is exhibited in Figure 1, the RSSI values from three APs fluctuate over a period of time. It
manifests that the RSSI value is not very stable. The RSSI data in Figure 1 is derived from a simple
experiment. Three APs are placed in the experimental space. The RSSI data was collected by a
trained person that stood at a fixed position, holding the mobile device and receiving signals from
Sensors 2019, 19, 2508 4 of 18

As is exhibited in Figure 1, the RSSI values from three APs fluctuate over a period of time. It
manifests that the RSSI value is not very stable. The RSSI data in Figure 1 is derived from a simple
experiment. Three APs are placed in the experimental space. The RSSI data was collected by a trained
person that stood at a fixed position, holding the mobile device and receiving signals from three APs
over a period of time. The distance between the three APs and the mobile device is different, resulting
in inconsistent strengths of the three groups of signals in Figure 1 where the orange signal comes from
the nearest AP while the green signal comes from the farthest AP.

• Multipath effect is the major factor affecting indoor localization accuracy because of signals
received from different paths [26].
• There is a difference in the orientation of the collector or user when collecting signals [25].

All of these factors affect the fingerprinting database and ultimately have impacts on
positioning accuracy.

2.2. Wi-Fi Fingerprinting Data Collection

Wi-Fi fingerprinting localization is also known as scene analysis. Commonly, scene analysis
consists of two phases, as presented in Figure 2, including the offline training phase and online
localization phase. In the offline training phase, WSNs sensor modules are placed in an exactly
indoor environment on fixed locations (access points). Then, a trained person stands at predefined
locations (reference points), holding the mobile device and collecting the RSSI values from APs. The
collection proceeds through a list of predefined locations. Finally, these collected RSSI values and
their corresponding coordinates compose a fingerprinting database or are called a radio map for
estimating the locations of target points, which include users with smartphone or robots in the WSNs
environment [27]. During the online localization phase, researchers collect the RSSI values of users
and match them to the fingerprinting data in the radio map by a localization algorithm to estimate the
user’s location. In the online localization phase, testing samples consist of the RSSI values collected by
the user at unknown target locations. For the estimation of the target locations, a localization algorithm
is used2019,
Sensors in matching
19, 2508 the testing samples to the fingerprinting data in the radio map [28]. 5 of 21

Figure 2. Two phases of fingerprinting

fingerprinting localization.
localization.

Besides, most of the studies are evaluated in a specific environment or on a private dataset
which hampers the reproducibility of the experimental results and has a negative impact on the
comparison between different algorithms [12,27,28]. For the purpose of foster comparison and
development in indoor localization research, some databases have been made available to the
Sensors 2019, 19, 2508 5 of 18

Besides, most of the studies are evaluated in a specific environment or on a private dataset which
hampers the reproducibility of the experimental results and has a negative impact on the comparison
between different algorithms [12,27,28]. For the purpose of foster comparison and development in
indoor localization research, some databases have been made available to the public. However, there
are two major problems with most of the current databases. Firstly, it is difficult to maintain the
database for a long time. Secondly, it is difficult to apply the database to new algorithms.
In order to handle the mentioned problems, a long-term Wi-Fi fingerprinting database for the
library of Jaume I University was selected for performance evaluation in this paper [29]. The origin
version of this database was founded on 24 November, 2017 and has been updated twice. The latest
version 2.1 was uploaded on 6 November, 2018 on Zenodo. The data collection spanned over 15 months
on version 1.0 and 25 months on version 2.1. Experiments in the paper are based on version 2.1.
Specifically, collecting data of each month from the database involved a training dataset and five
groups of test datasets. According to Figure 3, the original training and test dataset are divided into
three datasets including new training dataset, new validation dataset and new test dataset. Firstly,
the original training dataset was considered as the new training dataset which would be adopted in
training a CNN model. Then, different from the usage in [19], the original test dataset was divided into
two parts including new validation dataset that consisted of the first group of original test datasets and
a new test dataset that consisted of other groups of original test datasets. The validation dataset would
be adopted in training a GPR model. The datasets that appear below refer to new datasets without an
Sensors 2019, 19, 2508 6 of 21
additional explanation.

Figure 3. Dataset reorganization.

Figure 3. Dataset reorganization.
2.3. Signals Pre-Processing
For thePre-processing
2.3. Signals purpose of training CNN, the data processing focused on transforming the data into a
desired form which would be successfully understood and utilized.
For the purpose of training CNN, the data processing focused on transforming the data into a
In order to improve training efficiency, the following steps were taken prior to transformation.
desired form which would be successfully understood and utilized.
Firstly, the RSSI values were set to zero if they were below −105 dBm and, secondly, all of the remaining
In order to improve training efficiency, the following steps were taken prior to transformation.
signals were normalized. As for the first step, signals value less than this threshold usually means that
Firstly, the RSSI values were set to zero if they were below −105 dBm and, secondly, all of the
the communication distance is too far to provide valuable information for localization. Meanwhile, the
remaining signals were normalized. As for the first step, signals value less than this threshold
Wi-Fi-based RSSI values are susceptible to noise interference so that signals below the threshold would
usually means that the communication distance is too far to provide valuable information for
be unreadable. As for the second step, min–max normalization was applied in the remaining signals as
localization. Meanwhile, the Wi-Fi-based RSSI values are susceptible to noise interference so that
signals below the threshold would be unreadable. xi − xAs
minfor the second step, min–max normalization
x = . (1)
was applied in the remaining signals as i
e
xmax − xmin

xi − xmin
xi = . (1)
xmax − xmin

Furthermore, the data collected during 25 months were reorganized according to the dataset
that they belong to instead of the month they were collected. As a result, the training dataset
contained coordinates of 24 RPs and 300 RSSI vectors, respectively. The validation dataset contained
Sensors 2019, 19, 2508 6 of 18

Furthermore, the data collected during 25 months were reorganized according to the dataset that
they belong to instead of the month they were collected. As a result, the training dataset contained
coordinates of 24 RPs and 300 RSSI vectors, respectively. The validation dataset contained coordinates
of 24 VPs and 300 RSSI vectors, respectively. The test dataset contained coordinates of 82 TPs and
300 RSSI vectors, respectively. Therefore, each point and its corresponding numbers of RSSI vectors
constitute the Wi-Fi fingerprinting database adopted in this study.
After the above pre-processing steps, normalized signal vectors were organized into a desired
form. The first 256 RSSI vectors of each point including all of RPs, VPs and TPs from each dataset
were shaped as a 16 × 16 matrix. Specifically, the selected 620-dimensional vectors were sequentially
arranged as a matrix with a height of 16, a width of 16 and a depth of 620. Moreover, the form of the
fingerprinting database were changed as coordinates of 24 RPs and 24 corresponding matrixes in the
training dataset, coordinates of 24 VPs and 24 corresponding matrixes in the validation dataset and
coordinates of 82 TPs and 82 corresponding matrixes in the test dataset.
In a typical CNN, the training data is usually a single-channel grayscale image or a three-channel
color image [30]. Similarly, the training data in this study can be considered as multi-channel “images”.
Different from the real images, the number of channels in the “images” in this paper reaches 620.

3. Proposed Model
Sensors 2019, 19, 2508 7 of 21

3.1. Model Structure

3.1. Model Structure
The structure
The structure of the model
of the model is illustrated
is illustrated in Figure
in Figure 4. 4.The
Thepre-processing
pre-processing database is divided
database is divided into
three parts:into three parts: training dataset, validation dataset and test dataset. They are adopted in training
training dataset, validation dataset and test dataset. They are adopted in training and
and testing CNN or GPR in different phases.
testing CNN or GPR in different phases.

Figure 4. The structure of the proposed model.

Figure 4. The structure of the proposed model.

In the first phase of the model, the training dataset was adopted in CNN training, then we got a
trained CNN model. In the second phase, we first employed the validation dataset to evaluate the
accuracy of the CNN model’s position prediction for VPs and to calculate the positioning error.
Then, we trained the GPR model by the RSSI vectors from the validation dataset and the positioning
Sensors 2019, 19, 2508 7 of 18

In the first phase of the model, the training dataset was adopted in CNN training, then we got a
trained CNN model. In the second phase, we first employed the validation dataset to evaluate the
accuracy of the CNN model’s position prediction for VPs and to calculate the positioning error. Then,
we trained the GPR model by the RSSI vectors from the validation dataset and the positioning error.
Finally, we utilized the trained GPR model to correct the coordinates of the CNN algorithm’s position
prediction of TPs for the purpose of improving the localization accuracy.
Many researchers have utilized CNN for image classification. A unified CNN-RNN framework
isSensors
utilized for multi-label image classification [31]. A hybrid approach combines deep CNN8and
2019, 19, 2508 of 21
unsupervised aggregators was proposed in image classification for reducing learning cost [32]. However,
few researchers
applicable employWi-Fi
to general the algorithm
fingerprintingfor indoor localization.
data. The In fact, CNN
model proposed in thisplays
papera unique
should role in
convert
studying
the generalindoorWi-Filocalization [33]. Existing
fingerprinting data into CNN-based
the required localization
form while algorithms
ensuring are performance
not applicableafter to
general Wi-Fi fingerprinting data. The model proposed in this paper
positioning. Furthermore, notwithstanding that the CNN model exhibits ideal localization should convert the general Wi-Fi
fingerprinting
performance,data the into the required
localization form while
model ensuring
integrating CNNperformance
and GPR after positioning.
presents betterFurthermore,
positioning
notwithstanding
performance than that the CNN
using the CNN model exhibitsalone.
algorithm ideal localization performance, the localization model
integrating CNN and GPR presents better positioning performance than using the CNN algorithm alone.
3.2. Convolutional Neural Network Architecture and Training
3.2. Convolutional Neural Network Architecture and Training
The proposed CNN model was employed as a probabilistic estimator. The network was
The proposed
organized CNN model
in a layered was employed
architecture as a probabilistic
as summarized in Figureestimator. The network
5. It has seven was organized
layers including three
inconvolutional
a layered architecture as summarized in Figure 5. It has seven layers
layers, three pooling layers and a fully-connected layer. The input layer including three convolutional
is the 16 × 16
layers,
× 620 three pooling
“images” from layers and a fully-connected
the training dataset. For the layer.
firstThe input layer islayer,
convolutional the 1616××1616× ×620 “images”
1000 feature
from the training dataset. For the first convolutional layer, 16 × 16 ×
maps are created by 1000 3 × 3 filters with a rectified linear unit (ReLU) activation function. 1000 feature maps are created
The
by 1000 3 ×
padding 3 filters
and stridewith a rectified
are both linear
set to one. Theunit (ReLU) activation
convolutional layer isfunction.
followedThe padding
by the and stride
functional layers.
are both set
Spatial to one. The
reduction layerconvolutional
is one of the layer is followed
functional by theMaximum
layers. functionalpooling
layers. Spatial
spatial reduction
reduction
layer
functional layer is selected to progressively reduce the spatial size, resulting in thelayer
is one of the functional layers. Maximum pooling spatial reduction functional is selected
reduction of the
tocomputational
progressively complexity
reduce the spatial size, resulting in the reduction of the computational
of the network and prevention of overfitting to some extent. Moreover, complexity
ofthe
thepooling
network andseparately
layer preventionshrinksof overfitting
each ofto thesome
input extent.
depthMoreover, the pooling
slices in space. The most layer separately
general form,
shrinks each of the input depth slices in space. The most general form,
2 × 2 (width and height) pooling window with a stride of two, is set to perform subsampling 2 × 2 (width and height) pooling
which
window
selects thewithmax
a stride
overoffourtwo,number
is set to and
performthensubsampling
shifts for two which selects
pixels. Afterthethat,
maxtheover four number
volume of each
and then shifts for two pixels. After that, the volume of each
“image” changes to 8 × 8 × 1000. The depth dimension of each “image” stays constant “image” changes to 8 × 8 × 1000.
duringThe the
depth
poolingdimension
process.of each “image” stays constant during the pooling process.

Figure5.5.Proposed
Figure Proposedconvolutional
convolutionalneural
neuralnetwork
network(CNN)
(CNN)model
modelarchitecture.
architecture.

For
Forthe
thesecond
secondconvolutional
convolutionallayer, 20003 3××33filters
layer,2000 filterswith
withReLU
ReLUactivation
activationfunction
functionareareapplied
applied
totocreate 8 × 8 ×
create 8 × 8 × 2000 feature maps. Similarly, another maximum pooling layer is adoptedininthe
2000 feature maps. Similarly, another maximum pooling layer is adopted the
meantime
meantimewhich
whichresizes
resizesthe
therepresentations
representationsas as44×× 44 ×× 2000. For the
2000. For the third
third convolutional layer, 44 ××44×
convolutional layer,
×4000
4000feature
featuremaps
maps are
are built
built by 4000 33 ×× 33 filters
by 4000 filterswith
withReLU
ReLUactivation
activationfunction.
function.Then,
Then,maximum
maximum
pooling layer with 2 × 2 pooling window reduces the volume to
pooling layer with 2 × 2 pooling window reduces the volume to 2 × 2 × 4000. 2 × 2 × 4000.
After
Afteraaseries
series of
of convolutional
convolutional and spatial reduction
and spatial reduction layers,
layers, aafully
fullyconnected
connectedlayerlayeroccurs.
occurs.A
Acombination
combinationofofconvolutional
convolutionallayers
layers with
with spatial
spatial reduction
reduction layers
layers work as an automaticinput
work as an automatic input
preprocessing unit that replaces the traditional complex feature extraction procedures.
preprocessing unit that replaces the traditional complex feature extraction procedures. They serve They serve as
as an automatic feature extraction layer providing features to the following fully connected neural
layer. A fully connected layer has all the neurons connected to all outputs of the last convolutional
layer. Then, a readout layer calculates the corresponding output.
Finally, a layer with softmax activation was employed to yield probabilities for each RP. For a
classifier, the maximum probability in this vector was selected to determine the classification of the
Sensors 2019, 19, 2508 8 of 18

an automatic feature extraction layer providing features to the following fully connected neural layer.
A fully connected layer has all the neurons connected to all outputs of the last convolutional layer.
Then, a readout layer calculates the corresponding output.
Finally, a layer with softmax activation was employed to yield probabilities for each RP. For a
classifier, the maximum probability in this vector was selected to determine the classification of the
image. As can be seen in Figure 5, the architecture of the CNN algorithm is shown above. In this paper,
the probabilities for each RP are considered as the weights for each reference coordinates. Therefore,
the estimated coordinate of the target point is the weighted sum of the reference coordinates.
Dropout was set to 0.5 for the convolutional layers and for the fully connected layer in order to
prevent over-fitting in the model. The loss function was set to minimize the categorical cross-entropy
using Adam optimizer. The training was performed with a batch size of 50 and 30 epochs.
The CNN model was built and evaluated in Python 3.7 using Pytorch package.

3.3. Error Correction Model Using GPR

Generally speaking, validation dataset is usually adopted as a fake test dataset to adjust the
hyper-parameters. Specifically, it was employed to train the GPR model in this paper. In an effort to
further prevent CNN from over-fitting the training data, the trained GPR model was applied to further
improve the localization accuracy. More specifically, some related steps will be shown as follows.
Prior to training a GPR model, the validation data was applied to evaluate the performance of
the trained CNN model. The localization error of the CNN model was obtained by calculating the
coordinate difference between the actual location and the predicted location. The training dataset
format required for the GPR model is different from the existing dataset format. Hence, a new training
dataset for GPR was built:

• Input data: 24 sets of 620-dimensional vectors from the validation dataset. Each set contains
16 × 16 signal vectors and 24 sets contain a total of 6144 signal vectors.
• Output data: 24 sets of two-dimensional vectors from the localization errors between the predicted
coordinates from the trained CNN and their real coordinates. Since each set of input data
corresponds to the same VP, the input data (signal vector) of the same set corresponds to the same
output data (error vector). Therefore, the output data includes a total of 6144 error vectors.

Following that, we proposed to utilize the GPR algorithm to establish the relationship between
the input data and the output data from the GPR training dataset. The input data is several high
latitude vectors which results in a complex nonlinear relationship between the input data and the
output data. GPR has the natural advantage of dealing with complex models. It is a non-parametric
algorithm which means that it can be expressed as any functional form, which makes it useful in
dealing with any complex models. Since training data is processed into a form that is suitable for CNN,
the training set becomes smaller. However, most machine learning algorithms need to rely on big data
to improve the accuracy of their predictions. Fortunately, the GPR algorithm is a good choice if the
application scenario is a highly nonlinear model and has a small training set. Therefore, we utilized
the non-parametric Gaussian process regression algorithm to build the error correction model.

3.4. GPR in Function-Space View

Generally, a Gaussian process is usually specified by its mean function and covariance function [34].
We defined the mean function µ(x) and the covariance function k(x, x0 ) of a real process f (x) as,

µ(x) = E[(x)]
(
, (2)
k(x, x0 ) = cov( f (x), f (x0 ))

which can be written as

f (x) ∼ GP(µ(x), k(x, x0 )) (3)
Sensors 2019, 19, 2508 9 of 18

A GP is defined as a collection of random variables. Usually, x, the index set of the random
variables is time, which represents that the study object is a set of sequences about time. However, the
index set x in this paper is RSSI, x ∈ Rd , d = 620, rather than time.
We consider our model in such a simple form

y = f (x) + ω. (4)

where f (x) ∼ GP(µ(x), k(x, x0 )) and ω ∼ N (0, σ2n ). ω is additive Gaussian noise with zero mean and
variance σ2n . The acquisition of σ2n will be demonstrated in Section 3.6. We have n = 6144 pairs of
training data D = (xi , yi ) ni=1 , where each input data xi ∈ Rd is a signal vector and each output data


yi ∈ R2 is a localization error vector. For notational convenience, we aggregate the n input vectors xi
into a d × n matrix X and output vectors yi into y. The GP defines posterior distributions over function
from input data to output data which indicates that [ f (x1 ), f (x2 ), f (x3 ), · · · , f (xn )] obeys a multivariate
Gaussian distribution.

[ f (x1 ), f (x2 ), f (x3 ), · · · , f (xn )]T ∼ N (M, Kij )





M = [(µ(x1 ), µ(x2 ), µ(x3 ), · · · , µ(xn ))]T

(5)




 Kij = k(xi , x j )


where M is the mean vector and K is the kernel function k(xi , x j ) of ith and jth element which is represented
by an n × n covariance matrix of inputs X. These distributions are represented non-parametrically. A key
idea underlying GP is the requirement that the function values at different points are correlated. We aim
to predict the function value (denoted as f∗ ) of testing RSSI vectors (denoted as X∗ = [x∗1 , x∗2 , · · · , x∗m ]),
conditioned on training data D. In our GPR model, training inputs and testing vectors have the same
dimension, which means m = n. The joint distribution of predictive object f∗ and the noise-free
observations in the training set can be obtained as

K(X, X) + σ2n I K ( X∗ , X ) T
" # " # " #!
y M(X )
∼N , (6)
f∗ M ( X∗ ) K ( X∗ , X ) K ( X∗ , X∗ )

where M(X) = [(µ(x1 ), µ(x2 ), · · · , µ(xn ))]T , M(X∗ ) = [(µ(x∗1 ), µ(x∗2 ), · · · , µ(x∗n ))]T , K(X, X) = Kij ,
Kij (X∗ , X) = k(x∗i , x j ) is the kernel function of ith and jth element, respectively, which is represented by
an m × n covariance matrix between testing vectors x∗i and training input while Kij (X∗ , X∗ ) = k(x∗i , x∗j )
is represented by an m × m covariance matrix of testing vectors. According to the theorem of
marginalization and conditional distribution, the predictive distribution of target can be obtained as

p( y∗ X, y, X∗ ) = N (M̂, Σ̂)




T 2 −1
 M̂ = K(X∗ , X) (K(X, X) + σn I ) ( y − M(X)) + M(X∗ )

(7)


 Σ̂ = K(X , X ) − K(X , X)T (K(X, X) + σ2 I )−1 K(X , X)


∗ ∗ ∗ n ∗

Taking noisy observations into account, the predictive distribution will be obtained from the
training set and testing RSSI vectors as

p( y∗ X, y, X∗ ) = N (µ̂, Σ̂ + σ2n I ). (8)

In addition, the predictive expression of mean function and variance function will be more concise
if we adjust the mean function of the training set to zero.

T −1
 M̂ = K(X∗ , X) (K(X, X) + σ2n I ) y


. (9)
 Σ̂ = K(X∗ , X∗ ) − K(X∗ , X)T (K(X, X) + σ2n I )−1 K(X∗ , X)


Sensors 2019, 19, 2508 10 of 18

The predictive distribution in (9) summarizes the key advantages of GP for the localization error
likelihood model. The GP posterior is estimated from acquired signal vectors to localization error
vector, assuming independence between different signal vector separately. During localization, the
likelihood of observing localization error vector can be computed at any signal vector using (9).

3.5. Kernel Function

The concept of kernel is central to GPR for predicting the testing targets. It is kernel that encodes
presumptions about the function that we hope to know. As for the presumptions, similarity between
input data x is the major problem due to closer input data being possible to have similar output data.
Therefore, it is feasible to utilize similarity between training points and TPs as a guidance about the
prediction of targets. We have selected three kernels that may be suitable for the RSSI vectors: Squared
Exponential (SE), Periodic (PER) and Matern [35].

3.5.1. Squared Exponential

The squared exponential is one of the most well-known kernel functions used for GPR,
as shown below
kx − x0 k
kSE (x, x0 ) = s2f exp(− ). (10)
2l2
where s2f is the signal variance and is also considered as an output-scale amplitude and parameter l is
the input length-scale that determines the strength of the correlation between inputs. The meaning of
these two parameters of the latter two kernel functions is the same as this kernel function. How these
parameters are derived from the training data D will be explained in Section 3.6. Before training the
GPR model, the variance is set to 1.0 and the length-scale is set to 0.2. The most striking feature of this
kernel function is that it is very smooth. One of the reasons is that it is infinitely differentiable. In other
words, the GP with SE kernel function can perform a mean-square derivation at any order.

3.5.2. Periodic
The periodic kernel is employed to model functions that exhibit a periodic pattern.
  0

 2 sin2 π (x−x ) 
 p 
kPER (x, x0 ) = kSE ($(x), $(x0 )) = s2f exp− . (11)
l2

 

where $(x) = [sin(πx/p), cos(πx/p)]T and p is the period. In the study, the variance is set to 1.0, the
length-scale is set to 0.2 and the period is set to 1.0. Signal vectors may appear periodic locally, so
periodic kernel function may be applicable to this model.

3.5.3. Matern Class

The Matern class of kernel function is given by
√ v  √ 
0 21−v  2v(x, x0 )   2v(x, x0 ) 
kMatern (x, x ) = s2f  Kv   (12)
Γ (v) 

l l 

with positive parameters v and l, where Kv is a modified Bessel function. It is possible that the most
interesting cases for machine learning are v = 3/2 and v = 5/2, for which
  √   √ 
0) = 1 + 3(x,x0 ) 3(x,x0 )
K ( x, x exp − l


 3/2 l


√  √ (13)
5(x,x0 )2
  
5(x,x0 ) 5(x,x0 )

 K5/2 (x, x0 ) = 1 +



l + 3l2 exp − l
Sensors 2019, 19, 2508 11 of 18

Regarding the setting of the parameter v, it is usually set between v = 1/2 and v = 7/2. If v ≤ 1/2,
the process will be too rough to get an ideal model. When there is no clear prior knowledge about the
existence of higher order derivatives, the limited noisy training examples will be the main problem if
we set v ≥ 7/2. After taking the above issues into account and comparing the commonly used values
v = 3/2 and v = 5/2, we set v = 3/2. At the same time, the variance was set to 1.0 and the length-scale
Sensors 2019, 19, 2508 13 of 21
is set to 0.2.
commonly used values v = 3 / 2 and v = 5 / 2 , we set v = 3 / 2 . At the same time, the variance was
3.6. Hyperparameter Estimation
set to 1.0 and the length-scale is set to 0.2.
We can set θ = (σ2n , s2f , l) as the hyperparameters. In periodic kernel, θ = (σ2n , s2f , l, p). The log
3.6. Hyperparameter Estimation
likelihood of the observations is given by
We can set θ = (σ n2 , s 2f , l ) as the hyperparameters. In periodic kernel, θ = (σ n2 , s 2f , l , p ) . The log
1 −1 1   n
− log pof( y|X,
likelihood θ) = ( y − is
the observations (X))Tby
Mgiven (K + σ2n I ) ( y − M(X)) + logK + σ2n I + log 2π, (14)
2 2 2
1 1 n
The process of − log p ( y | X , θ )the
obtaining = （ y − M ( X )) ( K + σ n I ) ( y − M ( X )) + log K into
T 2 −1 2
+ σ n I a+process
log 2π , of obtaining
(14) the
2 hyperparameter can be transformed 2 2
minimum value of (14). Such a process can be implemented by utilizing conjugate gradient descent.
The process of obtaining the hyperparameter can be transformed into a process of obtaining
Specifically, we can compute the partial derivatives of (14).
the minimum value of (14). Such a process can be implemented by utilizing conjugate gradient
descent. Specifically, we can compute the partial derivatives
of (14).
∂ −1 T ∂K
!
1 
−1

log p( y|X, θ ) = tr K y K y (15)
∂θ j ∂ 12  ∂θj
T ∂K
log p ( y X ,θ ) = tr  ( K −1 y )( K −1 y )  (15)
∂θ j 2  ∂θ j 
Next, the corresponding partial derivatives are calculated according to different kernel functions.
Next,
For example, thethe corresponding
partial derivatives partial
of each derivatives
element for are calculated
squared according
exponential to function
kernel differentiskernel
as follows
functions. For example, the partial derivatives of each element for squared exponential kernel
 ∂K
 2 = 2σn δ
function is as follows
∂σn



0 2
  ∂K = s2 exp∂K− 1 x−x0 2 (x−x )

    

  ∂l f  = 2σ
2 l δ l 3 (16)

  ∂K ∂σ 2  n 2 
2 expn − 1 x−x0

  ∂s2 = 2s


2
 1  x − x '   ( x − x ')
l 2
f 2

 ∂Kf
 = s 2f exp  −    (16)
∂l  2  l   l3

where δ is one if x and x in the kernel
0 function  are the same input and zero otherwise.

 ∂K 2
 1  x − x ' 2 
= 2 s f exp  − 

4. Experiments Results and Discussion ∂s 2f  2  l  
  

4.1. CNN δ is one

Training
where if x and x ' in the kernel function are the same input and zero otherwise.
Results
Figure 6 displays
4. Experiments an overview
Results of the CNN training process. It is clear that the loss is below
and Discussion
0.3 even after the 25th epoch.
4.1. CNN
The epochTraining
number Results
describes the number of times that the algorithm will be trained by the entire
training samples.
Figure 6 displays anwhich
The loss, means
overview theCNN
of the cross-entropy loss, isItutilized
training process. to measure
is clear that the loss isthe error0.3at the
below
softmax
evenactivation layer.
after the 25th The closer the loss is to zero, the more perfect the model.
epoch.

Figure
Figure 6. 6. Losscurve
Loss curveof
of CNN
CNN with
withthe
the30th
30thepoch.
epoch.
Sensors 2019, 19, 2508 12 of 18

The epoch number is set to 30. It takes approximately 22.5 min totally and 45 s per epoch to finish the
CNN training on a laptop equipped with 8 GB RAM and an Intel Core i5-6200 U CPU with 2.3 GHz clock.

4.2. GPR Training Results

For the purpose of applying a GPR model, GPy, from the Sheffield machine learning group,
was utilized to design a GP framework in python. GPy 1.9.6 packet was selected to evaluate the
performance of the GPR model with three different kernel functions. Prior to training the GPR model,
some related parameters are presented in Section 4.3. These previous parameters were continually
adjusted during the training and the final parameters will be used to describe the trained model.
The changes in the parameters of the three kernel functions before and after training are illustrated
in Table 1.

Table 1. Comparison of Squared Exponential (SE) parameters before and after training.

Kernel Functions Before/After Training Variance Length-Scale Gaussian Noise Periodic

Before training 1.0 0.2 1.0
SE
After training 5.0593 0.7013 0.5440 × 10−2
Before training 1.0 0.2 1.0 1.0
Periodic (PER)
After training 11.91 3.638 0.7328 × 10−6 0.9371
Before training 1.0 0.2 1.0
Matern
After training 9.966 1.619 0.2074 × 10−4
The parameters after the training round four significant figures.

For GP, the log marginal likelihood of the model (14) is seen as the objective function of the model
being optimized. The variance in Table 1 is the signal variance and is also considered as an output-scale
amplitude and the length-scale in Table 1 is the input length-scale that determines the degree of
correlation between inputs. Both parameters are factors that determine the smoothness of the functions
estimated by a GPR. After training, variance and length-scale of both methods increase significantly.
These indicate that the GPR model after training is more adaptive to the volatility of the RSSI signal
and the correlation between inputs is stronger. The specific estimation process of hyperparameters
is shown in Section 3.6. When the objective function of different kernel function models reaches the
minimum value, each hyperparameter element in the hyperparameter vector is optimal.
Before comparing the localization accuracy of GPR models based on different kernel functions, let
us take a look at the changes of the objective function values before and after the model training.
Table 2 provides a comparison of objective function values for the three kernel functions. What
stands out in Table 2 is that the GPR model with Matern kernel function has the lowest objective
function value after training. Meanwhile, the model with SE kernel has the second lowest value and
the model with PER kernel has the highest value. What is striking about the values in Table 2 is how
well the three models with different kernels fit the training data.

Table 2. Comparison of the objective function values for the three kernel functions.

Before Training After Training

SE kernel function 38.08 27.76
PER kernel function 38.22 28.69
Matern kernel function 39.45 27.34
The objective function values in Table 2 round four significant figures.

4.3. Localization Accuracy

We will demonstrate the localization cumulative error distribution (CED) of five different
algorithms including CNN, CNN and GPR with SE kernel function, CNN and GPR with PER kernel
 Sensors 2019, 19, 2508 16 of 21
Sensors 2019, 19, 2508 13 of 18
4.3. Localization Accuracy
We will demonstrate the localization cumulative error distribution (CED) of five different
function, CNN and GPR with Matern kernel function and KNN algorithm separately and compare
algorithms including CNN, CNN and GPR with SE kernel function, CNN and GPR with PER
their
kernel function,effects.
localization CNN and TheGPR KNNwithalgorithm
Materniskernel
one offunction
the algorithms
and KNN with the highest
algorithm localization
separately and
accuracy in [29]. The literature only gives the comparison of the 75th percentile
compare their localization effects. The KNN algorithm is one of the algorithms with the highest error between five
algorithms among 15 months on version 1.0 database, however it does not show
localization accuracy in [29]. The literature only gives the comparison of the 75th percentile error more detail. Therefore,
webetween
reproduced the KNN algorithm
five algorithms based on
among 15 months on the version
version 2.1 database
1.0 database, as a itreference
however does not forshow precision
more
comparison. As the setting
detail. Therefore, in [29], wethe
we reproduced setKNN
k to 9.algorithm based on the version 2.1 database as a
From the
reference forcurves demonstrated
precision comparison.inAs Figure 7, it isin
the setting apparent
[29], wethat
set kthe CNN and GPR combined model
to 9.
exerts significant influence
From the curves on the localization
demonstrated in Figure accuracy. Firstly, compared
7, it is apparent that the CNNto theand
KNN GPR algorithm,
combinedthe
CNNmodel exerts has
algorithm significant influence
a significant on the localization
improvement in positioningaccuracy. Firstly,
accuracy. compared90%
Specifically, to localization
the KNN
algorithm,
error the CNNCNN
of the proposed algorithm has aissignificant
algorithm improvement
less than 2.25 m, yet of thein positioning
KNN algorithm accuracy. Specifically,
is less than 4.80 m.
90% localization
Moreover, error of error
all localization the proposed CNN algorithm
of the proposed is less thanare
CNN algorithm 2.25less
m, than
yet of3.80
the KNN algorithmthe
m. Secondly,
is less than 4.80 m. Moreover, all localization error of the proposed
addition of the GPR model makes the localization accuracy of the improved CNN algorithm CNN algorithm are less than
even
3.80 m.
greater. Secondly, the
Accurately, 90%addition of theerror
localization GPRof model makes the localization
the CNN+GPR model is less accuracy
than 2.00of the improved
m while of the
CNN algorithm
proposed even greater.
CNN algorithm is less Accurately,
than 2.25 m.90% localization
In other words, error of the CNN+GPR
all localization model
error of the is less
CNN+GPR
than 2.00 m while of the proposed CNN algorithm is less than 2.25 m. In other words, all
model is less than 3.60 m.
localization error of the CNN+GPR model is less than 3.60 m.

(a) (b)
Figure
Figure 7. Comparisonsofoflocalization
7. Comparisons localizationCED
CEDconcerning
concerning CNN
CNN algorithm
algorithm including:
including:(a)
(a)Comparison
Comparison
between
between CNNCNN and
and KNN;
KNN; (b)(b)Comparison
Comparisonbetween
betweenCNN
CNN and
and CNN+GPR.
CNN+GPR.

From
From Figure
Figure 8, it8,can
it can be seen
be seen that that models
models with with different
different kernelkernel functions
functions both positive
both have have positive
impacts
impacts on positioning but differ in detail. From a qualitative perspective, the model with
on positioning but differ in detail. From a qualitative perspective, the model with Matern kernel function Matern
kernel function
performs performs best
best in localization in localization
performance whileperformance whilefunction
with PER kernel with PERperforms
kernel function performs
the worst.
Sensors 2019, 19, 2508 17 of 21
the worst.

Figure
Figure 8. Comparison
8. Comparison of localizationCED
of localization CEDfor
for models
models with
withthree
threedifferent
differentkernel functions.
kernel functions.

From a qualitative perspective, we can see the overall localization effects of the five algorithms
from Figure 9. Particularly, the three combined algorithms, based on the CNN algorithm, employ
GPR model to correct the error and achieve significant results. Meanwhile, CNN algorithm also
achieves an ideal result. From a quantitative perspective, we will find more details of localization
accuracy for various algorithms through mean absolute errors (MAEs) and 75th percentile errors in
Table 3.
 Figure 8. Comparison of localization CED for models with three different kernel functions.
Sensors 2019, 19, 2508 14 of 18

From a qualitative perspective, we can see the overall localization effects of the five algorithms
fromFrom
Figure 9. Particularly,
a qualitative the three
perspective, wecombined algorithms,
can see the based on effects
overall localization the CNN algorithm,
of the employ
five algorithms
GPR Figure
from model9.toParticularly,
correct thethe
error and
three achieve algorithms,
combined significant based
results.
onMeanwhile, CNN algorithm
the CNN algorithm, employ GPRalso
achieves
model an ideal
to correct theresult. From
error and a quantitative
achieve perspective,
significant we will find
results. Meanwhile, CNNmore detailsalso
algorithm of achieves
localization
an
accuracy
ideal for From
result. various algorithms through
a quantitative meanwe
perspective, absolute errors
will find more(MAEs)
detailsand 75th percentile
of localization errorsfor
accuracy in
Table 3.algorithms through mean absolute errors (MAEs) and 75th percentile errors in Table 3.
various

Figure 9. 9.Comparison
Figure Comparisonofoflocalization
localizationcumulative
cumulative error
error distribution (CED)
(CED) for
forfive
fivedifferent
differentalgorithms.
algorithms.

Table 3. The mean absolute errors (MAEs) and 75th percentile errors for different algorithms.
Table 3. The mean absolute errors (MAEs) and 75th percentile errors for different algorithms.
Algorithm MAE (m) 75th Percentile Error (m) Improvement
MAE 75th Percentile Error
1 Algorithm
CNN 1.3910 1.9188 Improvement
45.8%
(m) (m)
1 CNN+GPR (Gaussian
CNN process 1.3910 1.9188 45.8%
2 0.9989 1.3125 62.9%
regression) with SE kernel
CNN+GPR (Gaussian process regression)
2 3 CNN+GPR with PER kernel 1.06450.9989 1.3125
1.4625 62.9%
58.7%
with SE kernel
3 4 CNN+GPR with
CNN+GPR withMatern kernel
PER kernel 0.95541.0645 1.2875
1.4625 63.7%
58.7%
4 CNN+GPR
Averagewith Matern kernel
of CNN+GPR 1.00630.9554 1.2875
1.3542 63.7%
61.8%
5 Average of CNN+GPR
k-nearest neighbor (KNN) (reference) 2.63381.0063 1.3542
3.5425 61.8%
0%
5 k-nearest neighbor
KNN ([29], (KNN)
version (reference)
1.0 database) 2.6338 3.5425
3.4060 0%
KNN ([29], version 1.0 database) 3.4060
What stands out in Table 3 is the different localization accuracy for five algorithms represented by
the MAEs and 75th percentile errors. The first four algorithms are the new algorithms proposed in
this paper. Besides, the fifth algorithm, KNN, is employed in literature [29]. One concern expressed
regarding the KNN algorithm in [29] is its evaluation that is based on version 1.0 rather than the latest
version database. To be fair, we reproduced the KNN algorithm based on the version 2.1 database
as the reference algorithm. Meanwhile, we exhibited the 75th percentile error of the KNN algorithm
in [29].
According to Table 3, a number of issues are identified. The CNN and GPR hybrid algorithm
has the best accuracy and the average localization accuracy of the three hybrid algorithms is 1.3542 m
which decreases 29.4% and 61.8% error as compared with the CNN algorithm and KNN algorithm.
Concentrating on the three hybrid algorithms, the algorithm with Matern kernel improves the localization
accuracy by 2.0% and 12.0% compared to the algorithm with SE and PER, respectively. Additionally, the
localization accuracy of the CNN algorithm is higher than the KNN algorithm by 45.8%.
Overall, the most striking finding is that both of the CNN algorithm and the hybrid algorithm
based on three different kernel functions greatly improve the localization accuracy compared with the
KNN algorithm which performs well in literature [29]. In addition, regarding the performance for the
 Additionally, the localization accuracy of the CNN algorithm is higher than the KNN algorithm by
45.8%.
Overall, the most striking finding is that both of the CNN algorithm and the hybrid algorithm
based on three different kernel functions greatly improve the localization accuracy compared with
the KNN
Sensors algorithm
2019, 19, 2508 which performs well in literature [29]. In addition, regarding the performance
15 of 18
for the three kernels in this study, the results support the idea that Matern kernel function performs
best in both the regression model and the localization test.
three kernels in this study, the results support the idea that Matern kernel function performs best in
both the regression Experiment
4.4. Supplementary model and the localization test.

4.4. Supplementary Experiment

As for the practical application of the hybrid model, there are two solutions to solve real-time
problems. First, the sample taken at a time can be copied into a 16 × 16 “image”. After experimental
As for the practical application of the hybrid model, there are two solutions to solve real-time
verification, this solution does not lead to a significant decrease in positioning accuracy. Second, the
problems. First, the sample taken at a time can be copied into a 16 × 16 “image”. After experimental
acquisition frequency can be increased to more than 100 Hz to provide enough samples [36]. The
verification, this solution does not lead to a significant decrease in positioning accuracy. Second, the
following will focus on the experimental verification of the first solution.
acquisition frequency can be increased to more than 100 Hz to provide enough samples [36]. The
When the target object moves quickly, its dwell time at a certain point is extremely short. This
following will focus on the experimental verification of the first solution.
may result in insufficient time to collect enough signal data and cause the model to be unavailable.
When the target object moves quickly, its dwell time at a certain point is extremely short. This
We have adjusted the data preprocessing to cope with this situation. In details, the samples
may result in insufficient time to collect enough signal data and cause the model to be unavailable. We
consisting of multiple time signals in the test dataset and validation dataset is changed to be copied
have adjusted the data preprocessing to cope with this situation. In details, the samples consisting of
from a single time signal. Since the signal at each moment can be expanded to one sample, the
multiple time signals in the test dataset and validation dataset is changed to be copied from a single
sample size of the test dataset and the validation dataset is greatly increased. As shown in Figure 10,
time signal. Since the signal at each moment can be expanded to one sample, the sample size of the
each test sample is converted to 256 new test samples. The samples in the training dataset remain
test dataset and the validation dataset is greatly increased. As shown in Figure 10, each test sample is
unchanged.
converted to 256 new test samples. The samples in the training dataset remain unchanged.

Figure 10. Samples conversion.

Figure 10. Samples conversion.
The experimental results of the solution are shown in Table 4.
The experimental results of the solution are shown in Table 4.
Table 4. Comparison of 75th percentile errors between supplementary experiment and
original experiment.

75th Percentile Rate of Change

Algorithm MAE (m)
Error (m) Compared to Table 3
1 CNN 1.4420 1.8350 −4.37%
2 CNN+GPR with SE kernel 1.0073 1.3250 +0.95%
3 CNN+GPR with PER kernel 1.2218 1.6750 +14.53%
4 CNN+GPR with Matern kernel 0.9512 1.2750 +0.97%
Average of CNN+GPR 1.0601 1.4250 +5.37%
5 KNN (reference) 2.6338 3.5425
KNN ([29], version 1.0 database) 3.4060

Table 4 highlights the comparison of 75th percentile errors between the supplementary experiment
and original experiment. From a qualitative perspective, the positioning performance exhibited by
all algorithms is close to the original experiment. The hybrid algorithm is superior to CNN. The
performance of the Matern kernel function is optimal in the hybrid algorithms. From a quantitative
Sensors 2019, 19, 2508 16 of 18

perspective, the positioning accuracy of most algorithms is very close to the original experiment. PER
kernel function is not very suitable for the new test samples, resulting in a certain degree of drop in the
accuracy of its hybrid algorithm.
From the data in Table 4, we can see that the localization precision of the same algorithm in the two
experiments is quite close. The results indicate that the first solution proposed in the first paragraph of
Section 4.4 does not lead to a significant decrease in localization precision. The above experiments
prove that the proposed algorithms have certain application value.

5. Conclusions
This paper has proposed a wireless indoor localization model using convolutional neural network
and Gaussian process regression. The most remarkable result to emerge from the data is that the
CNN and GPR hybrid model improves the positioning precision by 61.8% in total compared to the
baseline algorithm. While CNN has improved the performance by 45.8%, the GPR algorithm further
enhanced the localization accuracy. This result has further strengthened our conviction that either the
hybrid model or the CNN model adapts to complex scenes and can be treated as a reference method in
complex indoor localization.

Author Contributions: Conceptualization, G.Z.; methodology, G.Z.; software, G.Z.; validation, G.Z.; formal
analysis, G.Z.; investigation, G.Z. and H.C.; resources, P.W.; data curation, G.Z.; writing—original draft preparation,
G.Z.; writing—review and editing, G.Z.; supervision, L.Z.; project administration, L.Z.
Funding: This research received no external funding.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Alletto, S.; Cucchiara, R.; Del Fiore, G.; Mainetti, L.; Mighali, V.; Patrono, L.; Serra, G. An indoor location-aware
system for an IoT-based smart museum. IEEE Internet Things J. 2016, 3, 244–253. [CrossRef]
2. Lin, K.; Chen, M.; Deng, J.; Hassan, M.M.; Fortino, G. Enhanced fingerprinting and trajectory prediction for
IoT localization in smart buildings. IEEE Trans. Automat. Sci. Eng. 2016, 13, 1294–1307. [CrossRef]
3. Song, Q.; Guo, S.; Liu, X.; Yang, Y. CSI amplitude fingerprinting-based NB-IoT indoor localization. IEEE
Internet Things J. 2018, 5, 1494–1504. [CrossRef]
4. Xiao, J.; Zhou, Z.; Yi, Y.; Ni, L.M. A survey on wireless indoor localization from the device perspective. ACM
Comput. Surv. 2016, 49, 25. [CrossRef]
5. Luo, R.C.; Hsiao, T.J. Dynamic wireless indoor localization incorporating with an autonomous mobile robot
based on an adaptive signal model fingerprinting approach. IEEE Trans. Ind. Electron. 2019, 66, 1940–1951.
[CrossRef]
6. Passafiume, M.; Maddio, S.; Cidronali, A. An improved approach for RSSI-based only calibration-free
real-time indoor localization on IEEE 802.11 and 802.15. 4 wireless networks. Sensors 2017, 17, 717. [CrossRef]
7. Khan, U.F.; Lazaridis, P.I.; Mohamed, H.; Albarracín, R.; Zaharis, Z.D.; Atkinson, R.C.; Tachtatzis, C.;
Glover, I.A. An efficient algorithm for partial discharge localization in high-voltage systems using received
signal strength. Sensors 2018, 18, 4000. [CrossRef]
8. Zhang, J.; Han, G.; Sun, N.; Shu, L. Path-loss-based fingerprint localization approach for location-based
services in indoor environments. IEEE Access 2017, 5, 13756–13769. [CrossRef]
9. He, S.; Chan, S.-H.G. Wi-Fi fingerprint-based indoor positioning: Recent advances and comparisons. IEEE
Commun. Surv. Tutor. 2016, 18, 466–490. [CrossRef]
10. Liu, W.; Fu, X.; Deng, Z.; Xu, L.; Jiao, J. Smallest enclosing circle-based fingerprint clustering and
modified-WKNN matching algorithm for indoor positioning. In Proceedings of the 2016 International
Conference on Indoor Positioning and Indoor Navigation (IPIN), Alcala de Henares, Spain, 4–7 October
2016; pp. 1–6.
11. Shin, B.; Lee, J.H.; Lee, T.; Kim, H.S. Enhanced weighted K-nearest neighbor algorithm for indoor Wi-Fi
positioning systems. In Proceedings of the 2012 8th International Conference on Computing Technology and
Information Management (NCM and ICNIT), Seoul, South Korea, 24–26 April 2012; pp. 574–577.
Sensors 2019, 19, 2508 17 of 18

12. Fang, X.; Jiang, Z.; Nan, L.; Chen, L. Optimal weighted K-nearest neighbour algorithm for wireless sensor
network fingerprint localisation in noisy environment. IET Commun. 2018, 12, 1171–1177. [CrossRef]
13. Abdou, A.S.; Aziem, M.A.; Aboshosha, A. An efficient indoor localization system based on affinity
propagation and support vector regression. In Proceedings of the 2016 Sixth International Conference on
Digital Information Processing and Communications (ICDIPC), Beirut, Lebanon, 21–23 April 2016; pp. 1–7.
14. Milioris, D.; Tzagkarakis, G.; Papakonstantinou, A.; Papadopouli, M.; Tsakalides, P. Low-dimensional
signal-strength fingerprint-based positioning in wireless LANs. Ad Hoc Netw. 2014, 12, 100–114. [CrossRef]
15. Wu, Z.; Xu, Q.; Li, J.; Fu, C.; Xuan, Q.; Xiang, Y. Passive indoor localization based on csi and naive bayes
classification. IEEE Trans. Syst. Man Cybern. Syst. 2018, 48, 1566–1577. [CrossRef]
16. Feng, C.; Au, W.S.A.; Valaee, S.; Tan, Z. Received-signal-strength-based indoor positioning using compressive
sensing. IEEE Trans. Mob. Comput. 2012, 11, 1983–1993. [CrossRef]
17. Mirowski, P.; Milioris, D.; Whiting, P.; Ho, T.K. Probabilistic radio-frequency fingerprinting and localization
on the run. Bell Labs Techn. J. 2014, 18, 111–133. [CrossRef]
18. Ferris, B.; Haehnel, D.; Fox, D. Gaussian processes for signal strength-based location estimation. In
Proceedings of the Robotics: Science and Systems, Philadelphia, PA, USA, 16–19 August 2006.
19. Xu, S.; Chou, W.; Dong, H. A robust indoor localization system integrating visual localization aided by
CNN-based image retrieval with Monte Carlo localization. Sensors 2019, 19, 249. [CrossRef] [PubMed]
20. Zhou, B.; Yang, J.; Li, Q. Smartphone-based activity recognition for indoor localization using a convolutional
neural network. Sensors 2019, 19, 621. [CrossRef] [PubMed]
21. Krizhevsky, A.; Sutskever, I.; Hinton, G.E. Imagenet classification with deep convolutional neural networks. In
Proceedings of the Advances in Neural Information Processing Systems, Lake Tahoe, NV, USA, 3–6 December
2012; pp. 1097–1105.
22. Farid, Z.; Nordin, R.; Ismail, M. Recent advances in wireless indoor localization techniques and system.
J. Comp. Netw. Communs. 2013, 2013, 12. [CrossRef]
23. Moghtadaiee, V.; Dempster, A.G. Design protocol and performance analysis of indoor fingerprinting
positioning systems. Phys. Commun. 2014, 13, 17–30. [CrossRef]
24. Garcia-Villalonga, S.; Perez-Navarro, A. Influence of human absorption of Wi-Fi signal in indoor positioning
with Wi-Fi fingerprinting. In Proceedings of the 2015 International Conference on Indoor Positioning and
Indoor Navigation (IPIN), Banff, AB, Canada, 13–16 October 2015; pp. 1–10.
25. Luo, J.; Zhan, X. Characterization of smart phone received signal strength indication for WLAN indoor
positioning accuracy improvement. J. Netw. 2014, 9, 739. [CrossRef]
26. Song, J.; Hur, S.; Park, Y.; Choi, J. An improved RSSI of geomagnetic field-based indoor positioning method
involving efficient database generation by building materials. In Proceedings of the 2016 International
Conference on Indoor Positioning and Indoor Navigation (IPIN), Alcala de Henares, Spain, 4–7 October
2016; pp. 1–8.
27. He, S.; Lin, W.; Chan, S.-H.G. Indoor localization and automatic fingerprint update with altered AP signals.
IEEE Trans. Mob. Comput. 2017, 16, 1897–1910. [CrossRef]
28. Jung, S.H.; Moon, B.-C.; Han, D. Performance evaluation of radio map construction methods for Wi-Fi
positioning systems. IEEE Trans. Intell. Transp. Syst. 2016, 18, 880–889. [CrossRef]
29. Mendoza-Silva, G.; Richter, P.; Torres-Sospedra, J.; Lohan, E.; Huerta, J. Long-term WiFi fingerprinting
dataset for research on robust indoor positioning. Data 2018, 3, 3. [CrossRef]
30. Iqbal, Z.; Luo, D.; Henry, P.; Kazemifar, S.; Rozario, T.; Yan, Y.; Westover, K.; Lu, W.; Nguyen, D.; Long, T.
Accurate real time localization tracking in a clinical environment using bluetooth low energy and deep
learning. PLoS ONE 2018, 13, e0205392. [CrossRef] [PubMed]
31. Wang, J.; Yang, Y.; Mao, J.; Huang, Z.; Huang, C.; Xu, W. Cnn-rnn: A unified framework for multi-label image
classification. In Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition,
Las Vegas, NV, USA, 27–30 June 2016; pp. 2285–2294.
32. Kulkarni, P.; Zepeda, J.; Jurie, F.; Perez, P.; Chevallier, L. Hybrid multi-layer deep CNN/aggregator feature
for image classification. In Proceedings of the 2015 IEEE International Conference on Acoustics, Speech and
Signal Processing (ICASSP), Brisbane, QLD, Australia, 19–24 April 2015; pp. 1379–1383.
33. Bregar, K.; Mohorčič, M. Improving indoor localization using convolutional neural networks on
computationally restricted devices. IEEE Access 2018, 6, 17429–17441. [CrossRef]
Sensors 2019, 19, 2508 18 of 18

34. Williams, C.K.; Rasmussen, C.E. Gaussian Processes for Machine Learning; MIT Press: Cambridge, MA, USA,
2006; pp. 63–71.
35. Chen, Z.; Wang, B.; Gorban, A.N. Multivariate gaussian and student −t process regression for multi-output
prediction. arXiv 2017, arXiv:1703.04455.
36. Skoglund, M.A.; Hendeby, G.; Nygårds, J.; Rantakokko, J.; Eriksson, G. Indoor localization using
multi-frequency RSS. In Proceedings of the 2016 IEEE/ION Position, Location and Navigation Symposium
(PLANS), Savannah, GA, USA, 11–14 April 2016; pp. 789–798.

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).