A Real-Time Pothole Detection Approach for Intelligent Transportation System

Research Article

A Real-Time Pothole Detection Approach for Intelligent

Transportation System

Hsiu-Wen Wang,

Chi-Hua Chen,

Ding-Yuan Cheng,

Chun-Hao Lin,

and Chi-Chun Lo

1_{Institute of Information Management, National Chiao Tung University, Hsinchu 300, Taiwan}

2_{Telecommunication Laboratories, Chunghwa Telecom Co., Ltd., Taoyuan 326, Taiwan}

3_{Department of Information Management, Hwa Hsia University of Technology, New Taipei 235, Taiwan}

Correspondence should be addressed to Chi-Hua Chen; [email protected] Received 14 August 2014; Revised 4 December 2014; Accepted 9 December 2014 Academic Editor: Jung-Fa Tsai

Copyright © 2015 Hsiu-Wen Wang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. In recent years, fast economic growth and rapid technology advance have led to significant impact on the quality of traditional transport system. Intelligent transportation system (ITS), which aims to improve the transport system, has become more and more popular. Furthermore, improving the safety of traffic is an important issue of ITS, and the pothole on the road causes serious harm to drivers’ safety. Therefore, drivers’ safety may be improved with the establishment of real-time pothole detection system for sharing the pothole information. Moreover, using the mobile device to detect potholes has been more popular in recent years. This approach can detect potholes with lower cost in a comprehensive environment. This study proposes a pothole detection method based on the mobile sensing. The accelerometer data is normalized by Euler angle computation and is adopted in the pothole detection algorithm to obtain the pothole information. Moreover, the spatial interpolation method is used to reduce the location errors from global positioning system (GPS) data. In experiments, the results show that the proposed approach can precisely detect potholes without false-positives, and the higher accuracy is performed by the proposed approach. Therefore, the proposed real-time pothole detection approach can be used to improve the safety of traffic for ITS.

1. Introduction

In recent years, fast economic growth and rapid technology advance have led to significant impact on the quality of traditional transport system. Intelligent transportation sys-tem (ITS), which aims to improve the transport syssys-tem, has become more and more popular. For the safety of traffic, road users often feel uncomfortable when they drive on the rough roads, especially the potholes on the road. In accordance with the statistics from the Ministry of Justice in Taiwan, the national compensation money is about 240

million dollars from 2008 to 2011 [1]. The pothole on the

road causes serious harm to drivers’ safety. Therefore, drivers’ safety may be improved with the establishment of real-time pothole detection system for sharing the pothole information. Moreover, more and more sensors which include G-sensors, electronic compass, gyroscope, global positioning system (GPS), microphone, and cameras are equipped in

mobile device (e.g., smartphone and iPad). Several applica-tions use these sensors in mobile devices and combine mobile sensing techniques to solve problems such as social network

[2], healthcare [3], environment monitoring [4], and traffic

information [5]. Therefore, using the mobile device based on

mobile sensing techniques to detect potholes is suitable and convenient.

This study proposes a pothole detection method based on the mobile sensing and shares the pothole information with road users and government. For this purpose, the mobile device should be equipped with G-sensors and GPS to collect accelerometer data and location information. The accelerometer data is normalized by Euler angle computation and is adopted in the pothole detection algorithm to obtain the pothole information. Moreover, the spatial interpolation method is used to reduce the location errors from GPS data. Then the pothole information is made public to improve the safety of traffic.

Volume 2015, Article ID 869627, 7 pages http://dx.doi.org/10.1155/2015/869627

The remainder of the paper is organized as follows.

Section 2presents and discusses the various techniques of image recognition method and mobile sensing method for

pothole detection. In Section 3, this study proposes a

real-time pothole detection method based on mobile sensing. This study also illustrates the experiment results and analyses in

Section 4. Finally, this study concludes the paper inSection 5.

2. Related Work

Some pothole detection methods have been proposed and can be classified into two groups: image recognition method and mobile sensing method. The advantages and short-comings of these methods are presented in the following subsections.

2.1. Image Recognition Method. Yu and Salari proposed a

pothole detection approach based on laser imaging tech-niques to collect road information. Then the artificial neural network algorithm (ANN) was used to analyze the road

information and detect potholes [6]. However, this approach

which requires a big computation power to recognize the laser images is unsuitable for mobile devices.

Lin and Liu used the support vector machine algorithm (SVM) to analyze images about road information for pothole

detection [7]. Although this approach can provide high

accuracy, a big computation power is required for image recognition. Therefore, this approach is also unsuitable for mobile devices.

2.2. Mobile Sensing Method. For BusNet project, the

G-sensor and GPS are equipped in the on-board unit (OBU) in bus to collect accelerometer data and location information. These data can be sent to data processing center via wireless networks, and data processing center can analyze these data to check whether the vectors of accelerometer data exceed the

thresholds for pothole detection [8]. However, this approach

requires that the batch accelerometer data is sent when bus enters the bus station. Therefore, this approach cannot provide real-time pothole detection information.

The pothole patrol system which was proposed by a project team from Massachusetts Institute of Technology

combined G-sensor and GPS. This system analyzed the

𝑥-axis accelerometer data and𝑧-axis accelerometer data and

equipped five data filters which include(1) speed, (2)

high-pass,(3) 𝑧-peak, (4) 𝑥𝑧-ratio, and (5) speed versus 𝑧 ratio

[9]. Although these data filters can detect potholes, only

𝑧-peak of data filter can obtain the precise pothole information.

However, high misjudgment of𝑧-peak of data filters with the

surge of road.

Nericell project used a smartphone based on Windows Mobile operation system which is equipped with G-sensor and GPS to collect and analyze accelerometer data for pothole

detection [10]. However, the smartphone in this project

should be equipped with the specific angle. Furthermore, this

project only considered analyzing𝑧-axis accelerometer data

with high misjudgment.

Mednis et al. proposed four pothole detection approaches

which include (1) Z-THRESH approach, (2) Z-DIFF

approach,(3) STDEV-Z approach, and (4) G-ZERO approach

to analyze the accelerometer data [11]. The accelerometer

data in this study is obtained from Tmote sensors, Texas

Instruments controllers, and Analog Devices G-sensors [12].

However, the results of Z-THRESH approach and G-ZERO approach would be influenced by peak value to generate more false-positives. The results of Z-DIFF approach and STDEV-Z approach are dependent on frequency and timing. Therefore, the design and comparisons of these approaches for mobile device require to be investigated.

2.3. Summary. In summary, due to the requirement of big

computation power for image recognition method, too many resources are allocated for this method to be an efficiency mobile device. Therefore, mobile sensing method is more suitable to detect potholes for mobile device. However, pre-vious pothole detection approaches based on mobile sensing only considered one threshold to detect pothole, and high false-positives are obtained from these approaches.

Conse-quently, this study considers Mednis’s approaches [11] and

proposes a real-time pothole detection method to improve the accuracy of pothole detection method.

3. Real-Time Pothole Detection Method

The proposed real-time pothole detection method based on

mobile sensing includes three steps:(1) accelerometer data

normalization, (2) pothole detection approaches, and (3)

pothole location determination.

3.1. Problem Definition. Some shortcomings are in previous

pothole detection methods as follows: (1) mobile device

should be equipped with the specific angle; (2) high

false-positives may be generated with considering only one

thresh-old for pothole detection;(3) the precise pothole location has

not been investigated.

Therefore, this study proposes a real-time pothole detec-tion method based on mobile sensing to collect and normal-ize the accelerometer data from mobile device for free angle establishment. Furthermore, a pothole detection algorithm is proposed to consider several thresholds and combine several pothole detection approaches for pothole detection accuracy improvement. Finally, the space interpolation method is adopted to determine pothole location for leaving shortcom-ings.

3.2. Accelerometer Data Normalization. For solving the

lim-itation of the specific angle in previous pothole detection approaches, this study uses Euler angle formulas to nor-malize the three-axis accelerometer data. The Euler angles describe the vector set in 3-dimensional Euclidean space three parameters and represent a sequence of three elemental

rotations. For example,Figure 1shows that the vector set of

accelerometer data is defined as{𝑥󸀠, 𝑦󸀠, 𝑧󸀠}. The 𝑥󸀠𝑦󸀠𝑧󸀠system

rotates about the𝑥󸀠-axis by angle𝛼. The 𝑦󸀠-axis is now at

angle𝛼 with respect to the 𝑦-axis, and the 𝑧󸀠-axis is now

z z󳰀 y y󳰀 𝛼 x = x󳰀

Figure 1: A case study of Euler angles (the𝑥󸀠𝑦󸀠𝑧󸀠 system rotates about the𝑥󸀠-axis by angle𝛼).

Euler angle formulas, the vector set{𝑥, 𝑦, 𝑧} can be calculated

by adopting the values of vector set{𝑥󸀠, 𝑦󸀠, 𝑧󸀠} and angle 𝛼

(shown in formulas(1)). Furthermore, the vector of each axis

can be calculated by using Euler angle formulas when the

system rotates about the𝑦󸀠-axis by angle𝛽 and 𝑧󸀠-axis by

angle𝛾 (shown in formulas(2)). Therefore, the vector of each

axis with 0 degree angle is referred to as baseline in this study.

In runtime stage, the vector set{𝑥, 𝑦, 𝑧} can be calculated

by adopting the vector set of baseline and rotation angle for accelerometer data normalization:

[𝑥 𝑦 𝑧] = [𝑥󸀠 _𝑦󸀠 _𝑧󸀠_{] [} [ 1 0 0 0 cos 𝛼 − sin 𝛼 0 sin 𝛼 cos 𝛼 ] ] , 𝑥 = 𝑥󸀠, 𝑦 = 𝑦󸀠× (cos 𝛼) + 𝑧󸀠× (sin 𝛼) , 𝑧 = 𝑦󸀠× (− sin 𝛼) + 𝑧󸀠× (cos 𝛼) , (1) [𝑥 𝑦 𝑧] = [𝑥󸀠 _𝑦󸀠 _𝑧󸀠_{] [} [ cos𝛽 0 sin 𝛽 0 1 0 − sin 𝛽 0 cos 𝛽 ] ] , 𝑥 = 𝑥󸀠× (cos 𝛽) + 𝑧󸀠× (− sin 𝛽) , 𝑦 = 𝑦󸀠, 𝑧 = 𝑥󸀠× (sin 𝛽) + 𝑧󸀠× (cos 𝛽) , [𝑥 𝑦 𝑧] = [𝑥󸀠 _𝑦󸀠 _𝑧󸀠_{] [} [ cos𝛾 − sin 𝛾 0 sin𝛾 cos 𝛾 0 0 0 1 ] ] , 𝑥 = 𝑥󸀠× (cos 𝛾) + 𝑦󸀠× (sin 𝛾) , 𝑦 = 𝑥󸀠× (− sin 𝛾) + 𝑦󸀠× (cos 𝛾) , 𝑧 = 𝑧󸀠_. (2)

3.3. Pothole Detection Approaches. This study considers the

previous four pothole detection approaches and proposes a pothole detection algorithm to combine these approaches for pothole detection improvement. The notations of these approaches are defined and summarized in Notations.

3.3.1. The First Pothole Detection Approach: Z-THRESH. The Z-THRESH approach considers the minimum value of

𝑧-axis accelerometer data as the threshold to detect pothole.

The value of𝑧-axis accelerometer data is about −9.80065 m/s2

when the G-sensor is laid horizontally. Moreover, the value

of 𝑧-axis accelerometer data is lower than −9.80065 m/s2

when the G-sensor is dropped off. Therefore, the 𝑧-axis

accelerometer data drops off quickly when the car enters

a pothole. Then 𝑧-axis accelerometer data increases when

the car leaves a pothole. Therefore, this study considers the

lowest value of𝑧-axis accelerometer data through a pothole in

experimental runs. Furthermore, the maximum value of the

minimum value of𝑧-axis accelerometer data in each run is

selected as the threshold𝜃₁which is suitable to detect pothole

for each experimental run (shown in formula(3)). In runtime

stage, the value of𝑓₁(𝑔_{𝑎,𝑖,𝑗}) is 1 when the value of 𝑔_{𝑎,𝑖,𝑗}is lower

than𝜃₁for pothole detection (shown in formula(4)).

The value of threshold is

𝜃₁= max 𝑎=1,1≤𝑖≤𝑛,𝑖∈𝑁𝑒𝑖≤𝑗≤𝑙min𝑖,𝑗∈𝑁 𝑔_{𝑎,𝑖,𝑗}. ₍₃₎ Detection function is 𝑓1(𝑔𝑎,𝑖,𝑗) = {1, if 𝑔_{0, others,}𝑎,𝑖,𝑗≤ 𝜃1 where𝑎 = 1, 1 ≤ 𝑖 ≤ 𝑛, 𝑖 ∈ 𝑁, 𝑗 ∈ 𝑁. (4)

3.3.2. The Second Pothole Detection Approach: Z-DIFF. The Z-DIFF approach considers the maximum difference of two

consecutive 𝑧-axis accelerometer records as the threshold

to detect pothole. Due to much decreasing and increasing

of𝑧-axis accelerometer data through a pothole, the velocity

of variation of𝑧-axis accelerometer data between time 𝑡_{𝑖,𝑗−1}

and time 𝑡_𝑖,𝑗 is calculated and used to detect pothole.

Therefore, this study retrieves the largest value of velocity of

variation of𝑧-axis accelerometer data through a pothole in

experimental runs. Furthermore, the minimum value of the maximum value of variation velocity in each run is selected

as the threshold 𝜃₂ which is suitable to detect pothole for

each experimental run (shown in formula(5)). In runtime

stage, the value of𝑓₂(𝑔_{𝑎,𝑖,𝑗}) is 1 when the value of |𝑔_{𝑎,𝑖,𝑗} −

𝑔_{𝑎,𝑖,𝑗−1}|/(𝑡_𝑖,𝑗 − 𝑡_{𝑖,𝑗−1}) is larger than 𝜃₂ for pothole detection

(shown in formula (6)). However, the limitation of this

approach is difficult to determine the time difference between 𝑡_{𝑖,𝑗−1}and𝑡_𝑖,𝑗, and the accuracy of this approach is influenced by this time difference.

The value of threshold is

𝜃₂= min

𝑎=1,1≤𝑖≤𝑛,𝑖∈𝑁𝑒𝑖≤𝑗≤𝑙max𝑖,𝑗∈𝑁 󵄨󵄨󵄨󵄨

󵄨𝑔𝑎,𝑖,𝑗− 𝑔𝑎,𝑖,𝑗−1󵄨󵄨󵄨󵄨_󵄨

Detection function is 𝑓₂(𝑔_{𝑎,𝑖,𝑗}) ={{_{{ { 1, if 󵄨󵄨󵄨󵄨󵄨𝑔𝑎,𝑖,𝑗_𝑡 − 𝑔𝑎,𝑖,𝑗−1󵄨󵄨󵄨󵄨󵄨 𝑖,𝑗− 𝑡𝑖,𝑗−1 ≥ 𝜃2 0, others, where 𝑎 = 1, 1 ≤ 𝑖 ≤ 𝑛, 𝑖 ∈ 𝑁, 𝑗 ≥ 1, 𝑗 ∈ 𝑁. (6)

3.3.3. The Third Pothole Detection Approach: STDEV(Z).

The STDEV(Z) approach considers the maximum standard

deviation of 𝑧-axis accelerometer data as the threshold to

detect pothole. Due to the perturbation motion of 𝑧-axis

accelerometer data through a pothole, the standard deviation

of𝑧-axis accelerometer data during 𝐾 records is calculated

and used to detect pothole. Therefore, this study gets the

max-imum value of standard deviation of 𝑧-axis accelerometer

data through a pothole in experimental runs. Furthermore, the minimum value of the maximum value of standard

deviation in each run is selected as the threshold𝜃₃which is

suitable to detect pothole for each experimental run (shown

in formula(7)). In runtime stage, the value of𝑓₃(𝑔_{𝑎,𝑖,𝑗}) is 1

when the value of√∑𝑗_{𝑘=𝑗−𝐾+1}(𝑔_{𝑎,𝑖,𝑘}− 𝜇_𝑖,𝑗)2/𝐾 is larger than

𝜃₃ for pothole detection (shown in formula (8)). However,

the limitation of this approach is difficult to determine the

value of𝐾 which means time period, and the accuracy of this

approach is influenced by this time period. The value of threshold is

𝜃3=_{𝑎=1,1≤𝑖≤𝑛,𝑖∈𝑁}min _𝑒 max 𝑖≤𝑗≤𝑙𝑖,𝑗∈𝑁 √ ∑𝑗𝑘=𝑗−𝐾+1(𝑔𝑎,𝑖,𝑘− 𝜇𝑖,𝑗)2 𝐾 , where𝜇_𝑖,𝑗=∑ 𝑗 𝑘=𝑗−𝐾+1𝑔𝑎,𝑖,𝑘 𝐾 . (7) Detection function is 𝑓₃(𝑔_{𝑎,𝑖,𝑗}) = { { { { { { { 1, if √∑ 𝑗 𝑘=𝑗−𝐾+1(𝑔𝑎,𝑖,𝑘− 𝜇𝑖,𝑗)2 𝐾 ≥ 𝜃3 0, others, where 𝑎 = 1, 1 ≤ 𝑖 ≤ 𝑛, 𝑖 ∈ 𝑁, 𝑗 ≥ 𝐾, 𝑗 ∈ 𝑁, 𝜇_𝑖,𝑗= ∑ 𝑗 𝑘=𝑗−𝐾+1𝑔𝑎,𝑖,𝑘 𝐾 . (8)

3.3.4. The Fourth Pothole Detection Approach: G-ZERO. The G-ZERO approach considers all three-axis accelerometer

data and selects a lower bound and upper bound to detect pothole for accuracy improvement. When a car passes through a pothole, all values of three-axis accelerometer data are near to zero. Therefore, this study considers the largest value of three-axis accelerometer data through a pothole as a candidate of lower bound and the lowest value of three-axis accelerometer data through a pothole as a candidate of upper bound. Then the minimum value of the maximum value of three-axis accelerometer data in each run is selected

as the lower bound 𝜃_4,1, and the maximum value of the

minimum value of three-axis accelerometer data in each run

is selected as the upper bound𝜃_4,2 (shown in formulas(9)

and(10)). In runtime stage, the value of𝑓₄(𝑔_{𝑎,𝑖,𝑗}) is 1 when

the value of𝑔_{𝑎,𝑖,𝑗}is larger than𝜃_4,1and is lower than𝜃_4,2for

pothole detection (shown in formula(11)).

The value of lower bound is

𝜃_4,1= min

𝑎∈{1,2,3},1≤𝑖≤𝑛,𝑖∈𝑁𝑒𝑖≤𝑗≤𝑙max𝑖,𝑗∈𝑁

𝑔_{𝑎,𝑖,𝑗}. ₍₉₎

The value of upper bound is

𝜃_4,2= max 𝑎∈{1,2,3},1≤𝑖≤𝑛,𝑖∈𝑁𝑒𝑖≤𝑗≤𝑙min𝑖,𝑗∈𝑁𝑔𝑎,𝑖,𝑗. (10) Detection function is 𝑓4(𝑔𝑎,𝑖,𝑗) = {1, if 𝜃_{0, others,}4,1≤ 𝑔𝑎,𝑖,𝑗≤ 𝜃4,2 where 𝑎 = {1, 2, 3} , 1 ≤ 𝑖 ≤ 𝑛, 𝑖 ∈ 𝑁, 𝑗 ∈ 𝑁. (11)

3.3.5. The Fifth Pothole Detection Approach: Proposed Approach. This subsection proposes a pothole detection

approach which combines and improves the Z-THRESH and G-ZERO approaches to detect pothole. Furthermore, the Z-DIFF and STDEV(Z) approaches are limited in accordance with time differences and time periods, so these two approaches are not adopted. The pseudocode of the proposed pothole detection approach is presented in

Algorithm 1. The input parameters of this proposed approach are three-axis accelerometer data, and the value of output is 1 when the proposed pothole detection approach supposes the car passed through a pothole. In the proposed approach, the parameter check method is used to record whether the value of𝑓₁(𝑔_{𝑎,𝑖,𝑗}) or 𝑓₄(𝑔_{𝑎,𝑖,𝑗}) is 1. When one of Z-THRESH and

G-ZERO approaches supposes that the car passed through

a pothole, the timestamp 𝑡_𝑖,𝑗 is recorded and compared

with the parameter check time. The value of output is 1 if

𝑡_𝑖,𝑗-𝑐ℎ𝑒𝑐𝑘 𝑡𝑖𝑚𝑒 is smaller than 𝜀 seconds, which means a

pothole is detected. Furthermore, the parameter check time can be trained and learned by historical data from each practical run.

3.4. Pothole Location Determination. For pothole location

determination, this study uses the space interpolation

method to obtain precise pothole location.Figure 2 shows

that two locations (i.e.,𝐿₁ and𝐿₂) and timestamps (i.e.,𝑡₁

and𝑡₂) obtained from GPS module are adopted in the space

interpolation method to determine the pothole location. The

function𝑑(𝐿₁, 𝐿₂) is defined as the distance between location

𝐿1and location𝐿2. Therefore, the pothole location𝐿𝑝can be

determined by using 𝑑 (𝐿₁, 𝐿_𝑝) =𝑑 (𝐿1, 𝐿2) × (𝑡2− 𝑡1) (𝑡₃− 𝑡₁) , 𝑑 (𝐿₂, 𝐿_𝑝) =𝑑 (𝐿1, 𝐿2) × (𝑡3− 𝑡2) (𝑡₃− 𝑡₁) . (12)

Input: 𝑔_{𝑎,𝑖,𝑗}, where𝑎 = {1, 2, 3}, 1 ≤ 𝑖 ≤ 𝑛, 𝑖 ∈ 𝑁, 𝑗 ∈ 𝑁

Output: The value of output is 1 when the proposed pothole detection approach

supposes the car passed through a pothole.

Set check method = 0 set check time = 0 while (𝑗 ∈ 𝑁)

if (𝑡_𝑖,𝑗-check time) > 𝜀 sec then

check method = 0 check time = 0

end if

if 𝑓1(𝑔1,𝑖,𝑗) = 1 then

if check method = 0 then

check method = 1 check time =𝑡_𝑖,𝑗

else if check method = 4 then return 1 else check time =𝑡_𝑖,𝑗 end if end if if 𝑓₄(𝑔_{𝑎,𝑖,𝑗}) = 1 then

if check method = 0 then

check method = 4 check time =𝑡_𝑖,𝑗

else if check method = 1 then return 1 else check time =𝑡_𝑖,𝑗 end if end if end while

Algorithm 1: The pseudocode of the proposed pothole detection approach.

d(L1, L2) × (t2− t1) (t3− t1) d(L1, L2) × (t3− t2) (t3− t1) GPS location L1at time t1 GPS location L2at time t3 at time t2 Pothole location Lp

Figure 2: The space interpolation method for pothole location determination.

4. Experimental Results

This section discusses the analyses of experimental results for accelerometer data normalization, pothole detection approaches, and pothole location determination.

4.1. The Analyses of Accelerometer Data Normalization. For

the analyses of accelerometer data normalization, this study

gives two case studies which include(1) the mobile device

with 0 degree angle as baseline (shown inFigure 3) and(2) the

mobile device with−15 degree angle (i.e., the value of 𝛼 is −15

inFigure 1) (shown inFigure 4). Then the𝑡-test and 𝐹-test are

Figure 3: A case study of the mobile device with 0 degree angle as baseline.

used to verify the difference between the𝑧-axis accelerometer

data of baseline in Case 1 and the𝑧-axis accelerometer data

after accelerometer data normalization in Case 2.

This study uses two-tailed𝑡-test to determine significance

of the difference between the mean of𝑧-axis accelerometer

data of baseline in Case 1 (𝜇₁= −9.8489) and the mean of

𝑧-axis accelerometer data after accelerometer data

normaliza-tion in Case 2 (𝜇2 = −9.8476). The sample sizes of Case 1

15∘

Figure 4: A case study of the mobile device with−15 degree angle.

determine significance of the difference between the variance

of 𝑧-axis accelerometer data of baseline in Case 1 (𝜎₁2 =

0.000077) and the variance of 𝑧-axis accelerometer data after

accelerometer data normalization in Case 2 (𝜎₂2= 0.000085).

Table 1shows the𝑡-test and 𝐹-test for the population means and variances of samples in Case 1 and Case 2. Experimental

results indicate that the null hypothesis (H0:𝜇₁ = 𝜇₂) in

𝑡-test is accepted, and another null hypothesis (H0:𝜎₁ = 𝜎₂) in

𝐹-test is also accepted. Therefore, using Euler angle formulas to normalize the accelerometer data from mobile device is suitable for free angle establishment.

4.2. The Accuracy of Pothole Detection Approach. For the

analyses of accuracy of pothole detection approach, this study selects a pothole (length: 58 cm; weight: 51 cm; and

depth: 6 cm) as a case study (shown in Figure 5) and 10

runs in experiment environment. The mean frequency of accelerometer data detection in G-sensor in mobile device is

124 counts/second. The𝑘-fold cross-validation [13] is used to

verify the accuracy of pothole detection approach. In

exper-iments, training and testing are performed 10 times (i.e.,𝑘 =

10). In iteration 𝑖, the accelerometer data in 𝑖th run therapy is selected as the test corpus, and the accelerometer data in other runs is collectively used to train the thresholds for

each approach.Table 2shows the comparisons of Z-THRESH

approach, Z-DIFF approach, STDEV(Z) approach, G-ZERO approach, and the proposed approach. The results show that the proposed approach can precisely detect potholes without false-positives and the accuracy of the proposed approach is 100%. Furthermore, this study also implemented and compared common machine learning methods which include ANN, SVM, and decision tree (DT) (shown in

Table 3). Although these machine learning methods can detect potholes, several false-positives are generated by them.

4.3. The Error of Pothole Location Determination. For the

analyses of error of pothole location determination, this study uses the accelerometer data and location information

from 10 runs inSection 4.2to verify the space interpolation

method. The results show that the error of pothole location determination is reduced from 17.47 meters to 11.74 meters after using the space interpolation method. Therefore, the space interpolation method is suitable to determinate the precise pothole location.

6 cm 51 cm

58 cm

Figure 5: A case study of a pothole in University Road, Hsinchu, Taiwan.

Table 1: The test results for accelerometer data with normalization. The value of𝑧-axis accelerometer data

mean (standard deviation) The 0 degree angle

(baseline) −9.8489 (0.000077)

The−15 degree angle after

normalization −9.8476 (0.000085)

Table 2: The false positive of each pothole detection approach. Detection

method 𝑍-THRESH 𝑍-DIFF STDEV(𝑍) 𝐺-ZERO Proposedmethod

False-positive 49 40 274 8 0

Table 3: The comparisons of different machine learning methods.

Detection method ANN SVM DT

False-positive 1626 255 282

5. Conclusions and Future Work

This study proposes a real-time pothole detection method based on the mobile sensing techniques. This method uses Euler angle computation to normalize the accelerometer data obtained from mobile device with free angle establishment. Moreover, a pothole detection approach is proposed to be combined with Z-THRESH and G-ZERO approaches for reducing the false-positives of pothole detection. Fur-thermore, the spatial interpolation method is adopted to obtain precisely the location of pothole. In experiments, the results show that the proposed approach can precisely detect potholes without false-positives and the accuracy of the proposed approach is 100%. Therefore, the proposed real-time pothole detection approach can be used to improve the safety of traffic for ITS.

However, the limitation of this study is sample size. In the future, more practical results will be retrieved and analyzed to deploy the proposed method everywhere. Furthermore, due to the limited battery capacity of mobile device, the issue about saving of computation power can be investigated. A green pothole detection approach is needed to reduce the frequency of accelerometer data detection with high accuracy of pothole detection.

Notations

𝑔_{𝑎,𝑖,𝑗}: The acceleration of the𝑎th axle of the 𝑗th

record in the𝑖th run

𝑡_𝑖,𝑗: The timestamp of the𝑗th record in the 𝑖th

run

𝑛: The number of runs

𝑒_𝑖: The𝑒_𝑖th entering pothole record in the𝑖th

run

𝑙_𝑖: The𝑙_𝑖th leaving pothole record in the𝑖th

run

𝑎: The coordinate axis of G-sensor in mobile

device (e.g., the value of𝑎 is 1 which

means𝑍-axis)

𝑓_𝑚(⋅): The output of the 𝑚th pothole detection

approach (e.g., the value of𝑓₁(⋅) is 1 when

the first pothole detection approach supposes that the car passed through a pothole)

𝐾: The third pothole detection approach

requires𝐾 records to calculate the

standard deviation

𝜃_𝑚: The value of threshold for the𝑚th

approach

𝜃_4,1: The value of lower bound for the fourth

pothole detection approach

𝜃_4,2: The value of upper bound for the fourth

pothole detection approach.

Conflict of Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgment

The research is supported by the National Science Council of Taiwan under Grants nos. NSC 102-2622-H-009-001-CC3, NSC102-2410-H146-002-MY2, NSC 102-2410-H-009-052-MY3, and MOST 103-2622-H-009-001-CC3.

References

[1] Ministry of Justice, National Compensation Statistics, and Statistics of Justice, 2012,http://www.moj.gov.tw/ct.asp?xItem =36988&CtNode=11625&mp=095.

[2] E. Miluzzo, N. D. Lane, K. Fodor et al., “Sensing meets mobile social networks: The design, implementation and evaluation of the CenceMe application,” in Proceedings of the 6th ACM

Conference on Embedded Networked Sensor Systems (SenSys ’08), pp. 337–350, Raleigh, Calif, USA, November 2008.

[3] C.-C. Lo, C.-H. Chen, D.-Y. Cheng, and H.-Y. Kung, “Ubiqui-tous healthcare service system with context-awareness capabil-ity: design and implementation,” Expert Systems with

Applica-tions, vol. 38, no. 4, pp. 4416–4436, 2011.

[4] C. I. Wu, H. Y. Kung, C. H. Chen, and L. C. Kuo, “An intelligent slope disaster prediction and monitoring system based on WSN and ANP,” Expert Systems with Applications, vol. 41, no. 10, pp. 4554–4562, 2014.

[5] C.-H. Chen, H.-C. Chang, C.-Y. Su, C.-C. Lo, and H.-F. Lin, “Traffic speed estimation based on normal location updates and call arrivals from cellular networks,” Simulation Modelling

Practice and Theory, vol. 35, no. 1, pp. 26–33, 2013.

[6] X. Yu and E. Salari, “Pavement pothole detection and severity measurement using laser imaging,” in Proceedings of the IEEE

International Conference on Electro/Information Technology (EIT ’11), pp. 1–5, Mankato, Minn, USA, May 2011.

[7] J. Lin and Y. Liu, “Potholes detection based on SVM in the pavement distress image,” in Proceedings of the 9th International

Symposium on Distributed Computing and Applications to Busi-ness, Engineering and Science (DCABES ’10), pp. 544–547, Hong

Kong, August 2010.

[8] K. de Zoysa, C. Keppitiyagama, G. P. Seneviratne, and W. W. A. T. Shihan, “A public transport system based sensor network for road surface condition monitoring,” in Proceedings of the

Workshop on Networked Systems for Developing Regions, Kyoto,

Japan, August 2007.

[9] J. Eriksson, L. Girod, B. Hull, R. Newton, S. Madden, and H. Balakrishnan, “The pothole patrol: using a mobile sensor network for road surface monitoring,” in Proceedings of the 6th

International Conference on Mobile Systems, Applications, and Services (MobiSys ’08), pp. 29–39, June 2008.

[10] P. Mohan, V. N. Padmanabhan, and R. Ramjee, “Nericell: rich monitoring of road and traffic conditions using mobile smartphones,” in Proceedings of the 6th ACM Conference on

Embedded Network Sensor Systems, Raleigh, NC, USA, 2008.

[11] A. Mednis, G. Strazdins, R. Zviedris, G. Kanonirs, and L. Selavo, “Real time pothole detection using Android smartphones with accelerometers,” in Proceedings of the International Conference

on Distributed Computing in Sensor Systems (DCOSS ’11), pp. 1–

6, IEEE, Barcelona, Spain, June 2011.

[12] R. Zviedris, A. Elsts, G. Strazdins, A. Mednis, and L. Selavo, “LynxNet: wild animal monitoring using sensor networks,” in

Proceedings of the 4th International Conference on Real-world Wireless Sensor Networks, Colombo, Sri Lanka, 2010.

[13] J. Han and M. Kamber, Data mining: Concepts and Techniques, Morgan Kaufmann, San Francisco, Calif, USA, 2006.

Submit your manuscripts at

http://www.hindawi.com

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014

Mathematics

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014 Mathematical Problems in Engineering

Hindawi Publishing Corporation http://www.hindawi.com

Differential Equations

International Journal of

Volume 2014

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014 Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014

Mathematical PhysicsAdvances in

Complex Analysis

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014

Optimization

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014

Combinatorics

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014

International Journal of

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014

Journal of Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014

Function Spaces

Abstract and Applied Analysis

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014 International Journal of Mathematics and Mathematical Sciences

Hindawi Publishing Corporation http://www.hindawi.com Volume 2014

The Scientific

World Journal

http://www.hindawi.com Volume 2014

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014

Discrete Dynamics in Nature and Society

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014 Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014

Discrete Mathematics

Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014 Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014

Stochastic Analysis