An Intelligent System to Detect, Avoid and

Maintain Potholes: A Graph Theoretic Approach

Shreyas Balakuntala∗ , Sandeep Venkatesh†
M S Ramaiah Institute of Technology, Bangalore, India 560-054
∗ [email protected], † [email protected]

Abstract—In this paper, we propose a conceptual framework dissatisfied with the state of their roads. Some of the solutions
arXiv:1305.5522v3 [cs.AI] 18 Sep 2013

where a centralized system detects and assists the driver to avoid include: establishing pothole hot-lines, holding contests to re-
potholes on roads. The system also identifies the potholes which port particularly bad potholes, and asking readers to contribute
are to be repaired immediately. The system we propose comprises
a laser sensor and pressure sensors in shock absorbers to detect pictures of potholes. We seek a more systematic approach
and quantify the intensity of a pothole, a centralized server which to the problem, but hope that this public interest may cause
maintains a database of locations of all the potholes which can volunteers to carry hardware in their cars.
be accessed by another unit inside the vehicle. A point to point The existing systems propose a way to detect potholes.
connection device is also installed in vehicles so that, when a But, the vehicles travelling on these damaged roads eventually
vehicle detects a new pothole, all the vehicles within a range of
20 meters are warned about the pothole. We propose an algorithm take damages. We try to reduce this problem by proposing a
which computes a route with least number of potholes which is novel conceptual framework where the detection, avoidance
nearest to the desired destination. If the destination is unknown, and maintenance of potholes are taken care of to a great
then the system will check for potholes in the current road stretch possible extent. Our system accounts for a better,faster and
and displays the level of damage. Also, the destination can be reliable method in detection, avoidance and maintenance of
added, removed or changed any time during the travel. We prove
that the algorithm returns an optimal path. potholes. Thus this architecture, when employed brings about
Index Terms—Pothole, Weight Multiset, Min-Weight Multiset, a radical change in existing roadways transportation system.
Routing, Quick WiFi
A. Related Work
I. I NTRODUCTION There are several vehicular sensing methods for pothole
detection. Some of these use accelerometers or lasers for
Millions of dollars are spent in maintaining and repairing
data acquisition. This section contains a short review of
potholes by municipalities around the world1 . A pothole refers
pothole detection and avoidance algorithms implemented.
to a shallow pit on a road’s surface, caused by activities like
Some techniques use image processing for the detection.
erosion, weather, traffic and some other factors. These anoma-
lies when accumulated in the transportation system, constitutes
Mertz et. al., in [1] uses service vehicles to detect road
to major problems. These problems, even though they appear
damages. The system consists of a structured light sensor
to be less significant at an individual level, constitute to major
and a camera mounted on vehicles that travel the roads on a
problems when taken in cumulative, collective and large scale
regular basis. It makes use of sensors and equipment already
manner. The problems constituted by these potholes result in
present on the vehicle, like GPS on transit buses. The data
low fuel economy, accidents, traffic coagulations etc, which
is collected from many vehicles, aggregated and analysed at
have an adverse impact on the economy of a country and day
a central location and the assessment results are displayed
to day life of citizens. It can be proved that the traffic system
interactively to facilitate road maintenance operations. Rode
can be bettered to a greater extent when these problems are
et. al., [2] proposed a pothole detection and warning system
checked and taken care of. Detecting Potholes on roads with
which is divided into three subsystems. First is sensing
the help of intelligent systems is a very well-studied problem.
subsystem which senses the potholes encountered by it,
Detecting and hence avoiding potholes may reduce the fuel
about which it did not have the prior information. Second
consumption, wear-tear and maintenance cost of a vehicle.
is communication subsystem which handles the information
Also, avoiding potholes increases road safety and indirectly
transfer between Wi-Fi Access Point and Mobile Node.
decreases the total travel time in some cases.The existing sys-
Third subsystem is the localization subsystem which analyzes
tems of pothole detection uses a centralized database approach.
the data received from Access Points and warns the driver
Road conditions are a matter of public concern that have
regarding the occurrence of potholes. De Zoysa et al., in [3]
engendered a number of responses from local organizations
proposes a method of detecting potholes where pothole data
1 http://www.bizjournals.com/denver/stories/2007/04/02/story1.html?page= and its GPS coordinates are stored in the sensor. This data
all stored is uploaded to the base station when the bus reaches
the station. The station acts as an central hub and in turn is
handed over to the buses that travel in that route.
An image processing approach was proposed by Ajit et.al.,
in [4] where the potholes are first photographed or recorded
using camera on a car. This image processes the road such
that the pothole appears as a distinct black colour. The images
can be filtered to remove unnecessary objects like other cars
using proper pattern matching. This efficiently detects the
presence of potholes on the roads. Jengo et. al., in [5] uses
hyperspectral imagery to characterise the quality of roads, the
approach is of semi-automated in nature.

Smartphones have been used for effective detection and

avoiding mechanisms. An algorithmic approach was proposed
Fig. 1. The proposed system with 3 components.
by Eriksson et. al., [6] in addition to static road sensing.It
describes the design, implementation, and experimental
evaluation of Pothole Patrol, a mobile sensor computing .., (uvn , wnuv )}
system to monitor and assess road surface conditions.
An approach where accelerometers in Android Smartphones , where, uvi is the arc-i from vertex u to vertex v and wiuv
are used to detect potholes [7]. The system describes a mobile is weight of that arc, for i ={1, 2, 3,.., n}. If two or more
sensing system for road irregularity detection using Android number of weights is equal then ties are broken arbitrarily.
OS based smart-phones. It maps the road system by using The multiset {w1uv , w2uv , w3uv , ..., wnuv } is also denoted
smartphones and does statistical analysis of the mapped data by the name Wuv . This Wuv is called the ’Weight-
using existing RoadMic Technology. Strazdins et. al., in [8] Multiset’.Here wuv is the min value of the multiset
test the feasibility of such Android Smartphones for pothole Wuv = {w1uv , w2uv , w3uv , ..., wnuv }. Clearly, wuv = w1uv ,
detection. as the multiset Wuv is already sorted. The collection of all wuv
in a directed multigraph forms a multiset which is denoted
The problem is well studied for autonomous robots too. by W and is called by Min-Weight Multiset of the Multigraph.
A detection-avoidance mechanism was introduced for the
navigational aid of autonomous vehicles [9]. In this paper, For simplicity, we denote wiuv as w(e), where e is the
they discuss a solution to detection and avoidance of simulated i-th arc between u and v.
potholes in the path of an autonomous vehicle operating in an 3) Pothole Identification Number: Each pothole will be
unstructured environment. An obstacle avoidance system was assigned a unique pothole identification number Id which will
developed for a custom-made autonomous navigational robotic be linked to its location i.e, the arc on which it is present.
vehicle (ANROV), based on an intelligent sensor network and Any transaction/ process for a pothole will be done through
fuzzy logic control [10]. this number.

B. Definitions C. Brief overview of the results

1) Street Network: A street network is a directed multi- We propose a method where the detection, avoidance and
graph Gd = (V, Ed ) where V is the set of nodes which maintenance of potholes are taken care of to a great possible
correspond to interconnecting points and Ed is the set of extent. We use a peer to peer communication, along with the
edges(arcs) which correspond to interconnecting lines that centralized database approach. The vehicle will be routed such
represent a system of streets for a given area. Here |V | = N that the potholes in the path will be minimum. The potholes are
where N is the number of nodes. The graph is directed to regularly reported to the municipal corporation based on the
account for multiple lanes and two-way streets between any 2 number of hits. Hence the system accounts for a better,faster
vertices. and reliable method in detection, avoidance and maintenance
2) Weight Multiset and Min-Weight Multiset of a Directed of potholes.
Multigraph: We borrow the idea of Weight Multiset and Min- II. O UR A PPROACH
Weight Multiset from Siddhartha Sankar Biswas et al. [11].
Suppose that there are n number of arcs from the vertex u The proposed system is a conceptual framework which can
to the vertex v in a directed multigraph Gd , where n is a be implemented for vehicles in metropolitan cities. The system
non-negative integer. Wuv is the set whose elements are the consists of three components namely Detection, Avoidance
arcs between vertex u and vertex v, keyed and sorted in non- and Maintenance of the potholes. Figure 1 is a block diagram
descending order by the value of their respective weights. of the complete system. The components are implemented
using a set of agent(s). Each component will be explained
∴ Wuv = {(uv1 , w1uv ), (uv2 , w2uv ), (uv3 , w3uv ), .. in detail in the following sections.
A. Detection
1) Pothole Detection With Lasers: We attach a laser detec-
tor and transmitting device to the vehicles’ under carriage. The
device utilises a laser to detect the depth of the pothole. A 3-D
laser scanner [1] operates by sweeping a laser across the scene
in two dimensions. At each pixel, the instrument measures the
time that it takes for a laser beam to leave the sensor, strike
a surface, and return. The sensors also provide an intensity
measurement at each pixel by measuring the energy of the
returned laser signal. Thus, a full sweep in two-dimensions
can provide both a depth map and an intensity image using
Laser Detection and Ranging mechanisms [12]. The depth and
location of a pothole are linked to a pothole identification Id.
2) Encryption and Decryption: A data encryption and Fig. 2. An instance where vehicles are connected to the server through a
decryption system securely geoencrypts data using location- wifi access point. Here a new pothole is detected and is updated to the server
by the rightmost and the leftmost vehicles.
dependant navigation signals. To prevent any brute-force attack
against the cryptographic key, we use largely time-independent
characteristics of the navigation. It includes a geoencryption 2) Transmission of Data to the Central Server: Each device
apparatus [13] that includes the data file, consisting of the is a mini-server that supports a set of functions and can
depth map and the intensity image, the location of the device process portions of the queries [16]. It is also connected
along with a random key in order to create a coded file. The to a central database which manages the pothole location
decryption decrypts the same using RSA algorithm [14]. It discovered along with its depth. The database will store a tuple
discards the random number and utilises the location, depth (Depth, Location, Intensity) of potholes detected and a pothole
map and intensity image for further processing. Id will be pointing this tuple. On querying, the database return
results immediately. They monitor the roads as a continuous
B. Avoidance phenomenon. The central database system accepts a query
1) Warning: Local Transmission of Data: To transfer data regarding the condition of the road. It produces a query
locally between the vehicles we use Cabernet [15]. Cabernet is execution plan for the query and executes this plan against
a system for delivering data to and from moving vehicles using the database containing the information about the location and
open 802.11 (WiFi) access points(AP) encountered opportunis- depths of potholes.The execution plan is the internal blueprint
tically during travel.Network connectivity in Cabernet is both for evaluating a query. The answer to the query is encrypted
fleeting (where access points are typically within range for a and then send to the requesting devices.Query processing takes
few seconds)and intermittent. The advantages of using Quick advantage of the computing capabilities at the devices in order
WiFi is that data transfers can occur in broadband speeds. to minimise reaction time. Figure 2 shows the instance where
As a car drives down the road, the on-board embedded all the vehicles are connected to each other and to the server
device repeatedly scans for, and attempts to associate with, through Cabernet.
open APs. It then attempts to establish end-to-end connectivity 3) Preprocessing of the Street Network: The street network
with a Cabernet enabled host, to retrieve or upload the data will be preprocessed and weights will be added to all the
obtained. When a vehicle moves, the connections are typically arcs. These weights are based on the depth of the potholes(
brief. The device associates with an AP and maintains the damage level of the arc) and the length of the arc which s the
connection until the access point went out of range. In the actual physical distance between the 2 vertices considered.
connection establishment process each step involves a request, The weights will be updated for every update of a pothole in
followed by a response. If no response is received within the server. Algorithm 1 is used for the preprocessing. Here
the specified time, the request is retransmitted. QuickWiFi the weight wiuv , of a given i-th arc e between any 2 vertices
attempts to associate with the first open access point it (u, v), is given by the product of the average damage of the
encounters as it scans through the wireless channels.Open arc and its length i.e.,
APs require the connection to be authenticated before the
wiuv = d × l
transmission can occur. In case the authentication is lost,
the process is restarted.Upon success, QuickWiFi explicitly . The average damage of an arc is given by
notifies applications that connection is available and the data Pa
is transmitted.We quickly discover that a connection is lost, if j=1 bj
d=
we have not seen any transmissions for 500 milliseconds, it a
is likely that the car has moved out of the range of the AP. where a is the number of potholes on the arc and bj is the
When this happens, we are rescanning again to transmit data depth of j-th pothole. We consider the product of d and l to
to other vehicles within the range. account for both the damage and the distance. For an optimal
path, both damage and the distance are to be minimum and Algorithm 2 Routing to Destination
hence the weight wiuv accounts for a better measure. Require: Weighted Multi-graph G0d = (V, Ed ) , Min-Weight
Multiset W , Source S and Destination D as input.
1: if Destination D = ∅ then
Algorithm 1 Preprocessing of the Street Network 2: Get the GPS data of the vehicle i.e, Arc e on which the
Require: Street Network Gd = (V, Ed ), Weight Multiset vehicle is travelling.
Wuv for all (u, v) ∈ V and Pothole Identification Id for 3: return The weight w(e).
all the potholes in the Database as input. 4: Repeat steps 2 and 3 until the destination is either Set
1: for all e ∈ Ed do or Reached for every node crossed.
2: for all Id in the database do 5: else
3: if Pothole with identification number Id is on arc e 6: Run GDA(G0d ,W ,S).
then 7: return The path between S and D obtained from GDA
4: Arc-damage sum, d0 (e) = d0 (e) + b(Id) where algorithm along with its weight.
b(Id) is the depth of the pothole on arc e with 8: if D is modified then
identification Id. 9: Repeat step 6 and 7.
5: end if 10: end if
6: end for 11: end if
7: end for
8: for all e between any (u, v) do
0
9: Average arc-damage, d(e) = d a(e) where a is the
number of potholes on arc e.
10: Weight of arc, w(e) = d(e) × l(e) where l(e) is the
length of the arc e obtained from the weight multiset
(w(e) = wiuv assuming e to be the i-th arc between u
and v).
11: end for
12: return Weighted Multi-graph G0d = (V, Ed ) of Street
network Gd .

4) Routing the Vehicle through Optimal Path: The device

in the vehicle will have a display unit, which shows the street Fig. 3. The proposed system as a whole.
network as a map to the driver. The system is flexible enough
that the destination can be set at the source, while driving
and in addition to it, the destination can be modified. If the
destination is set at the source, a path which has least weight C. Maintenance: Determining Traffic Intensity using server
will be displayed. If its not set, then the weight of the arc updates for a particular pothole
on which the vehicle is travelling will be displayed. If the
The location of every pothole is updated by every vehicle
destination is modified, i.e., changed or deleted, one of the
which communicates the central the database. We utilise this
above 2 processes will be called. Algorithm2 will be followed
redundant effectively to find the traffic intensity of the roads.
for routing. In Algorithm 2, we use Generalised Dijkstra’s
Traffic Intensity is equal to the number of traffic updates
algorithm(GDA) for multi-graphs [11] to route the vehicle.
for a particular location per minute. The traffic intensity on
This GDA algorithm returns shortest path from a source to all
a particular location is compared by the number of pothole
other vertices.
updates received for a particular location in one minute.
Theorem 1: Algorithm 2 always returns an optimal path. This gives an approximate idea of the traffic intensity in the
Proof: The weights to the multi-graph G0d = (V, Ed ) is damaged roads. This data is properly exploited to prioritise
added such that both the length of an arc l(e) and its average the potholes on the level of threat to traffic intensive roads.
depth d(e) are minimal. This results in the weight of an arc The classified roads can be intimated to the corresponding
w(e) = l(e) × d(e) to be minimum. Generalized Dijkstras authorities for action to be taken upon.
algorithm returns a path which is the shortest between given Figure 3 shows the complete system where a tuple (Depth,
source-destination pair (S, D). The path is said to be optimal Location, Intensity) will be returned by the system upon
since it bears a weight which accounts for less damage and a detection of a pothole to the server. This tuple for a pothole is
shorter distance. Hence Algorithm 2 returns an optimal path used by other vehicle for avoidance and maintenance purpose.
always.
 III. C ONCLUSION [8] Strazdins, G., Mednis, A., Kanonirs, G., Zviedris, R. and Selavo, L.
(2011). Towards Vehicular Sensor Networks with Android Smartphones
This paper describes laser-based detection and avoidance for Road Surface Monitoring.
of potholes using routing algorithms and road damage level [9] Jaiganesh Karuppuswamy ; Vishnuvardhanaraj Selvaraj ; Meyyappa M.
Ganesh and Ernest L. Hall, ”Detection and avoidance of simulated
predictions. It also prioritises the potholes in important roads potholes in autonomous vehicle navigation in an unstructured environ-
based on traffic intensity. The method we use exploits the ment”, Proc. SPIE 4197, Intelligent Robots and Computer Vision XIX:
current location of the device and sends it to other vehicles Algorithms, Techniques, and Active Vision, 70 (October 11, 2000);
doi:10.1117/12.403788; http://dx.doi.org/10.1117/12.403788
using Quick Wifi APs. The same data is also forwarded to the [10] Chan Yet, W.; Qidwai, U., ”Intelligent sensor network for obstacle
central server. Thereby increasing the accuracy of the exact avoidance strategy,” Sensors, 2005 IEEE , vol., no., pp.4 pp.,, Oct. 30
location of the pothole. The road damage level predicting 2005-Nov. 3 2005 doi: 10.1109/ICSENS.2005.1597721
[11] Biswas, S.S., B. Alam and M.N. Doja, 2013. Generalization of Dijkstras
algorithm gives a scale of damage for the road ahead. The algorithm for extraction of shortest paths in directed multigraphs. J.
routing algorithm returns an optimal path to the destination Comput. Sci., 9: 377-382.”
with minimal damage and a shorter distance. [12] Yu, Miao, Jian Hua Ren, Dashsiung Hsu, and Tong Gang Zhao. ”High
Resolution LADAR Based on Scanning Distance-Imaging.” Advanced
Materials Research 459 (2012): 411-415.
IV. F URTHER W ORK [13] U.S Patent No: US 8,391,488 B2 Method and Apparatus for using
The data in the central server can be exploited by concerned navigation signal information for geoencryption to enhance security. Di
Qui, Menlo Park, CA(US)
authorities for remedial action on the potholes. The verification [14] R. L. Rivest, A. Shamir, and L. Adleman. 1978. A method for
of the remedial action can also be taken by detecting the obtaining digital signatures and public-key cryptosystems. Commun.
presence of potholes in the particular location. Also the ACM 21, 2 (February 1978), 120-126. DOI=10.1145/359340.359342
http://doi.acm.org/10.1145/359340.359342
efficiency of interaction of vehicles through Quick Wifi can be [15] Jakob Eriksson, Hari Balakrishnan, Samuel Madden, In Proceedings
improved to reduce data losses. The routing for avoidance can of the 14th ACM international conference on Mobile computing and
be studied in detail and efficient algorithms can be proposed. networking (2008), pp. 199-210, doi:10.1145/1409944.1409968
[16] Bonnet, P.; Gehrke, J.; Seshadri, P., ”Querying the physical world,”
Moreover, the system is still a theoretical proposition which Personal Communications, IEEE , vol.7, no.5, pp.10,15, Oct 2000 doi:
can be implemented in metropolitan cities. 10.1109/98.878531

V. ACKNOWLEDGEMENT
The authors would like to thank M.S.Ramaiah Institute
of Technology, Prof.K.G.Srinivasa( Head, Dept of CSE,
M.S.Ramaiah Institute of Technology) the support and mo-
tivation. The authors would also like to thank Dr. Ranjini
Tolakanahalli( Medical Physicist, Hamilton Health Sciences)
and Mr. Balakunatala Prasanna( Chair, IEEE Region 1) for the
review and suggestions.

R EFERENCES
[1] Christoph Mertz, Continuous Road Damage Detection Using Regular
Service Vehicles in Proceedings of the ITS World Congress, October
2011
[2] Rode, S.S.; Vijay, S.; Goyal, P.; Kulkarni, P.; Arya, K., ”Pothole Detection
and Warning System: Infrastructure Support and System Design,” Elec-
tronic Computer Technology, 2009 International Conference on , vol.,
no., pp.286,290, 20-22 Feb. 2009 doi: 10.1109/ICECT.2009.152
[3] De Zoysa, Kasun, Nayanajith M. Laxaman, and Kenneth M. Thi-
lakarathna. ”Automated Pothole Detection Using Wireless Sensor Motes.”
[4] Ajit Danti, Jyoti Y. Kulkarni, and P. S. Hiremath, ”An Image Processing
Approach to Detect Lanes, Pot Holes and Recognize Road Signs in Indian
Roads,” International Journal of Modeling and Optimization vol. 2, no.
6, pp. 658-662, 2012.
[5] Jengo, Christopher M., David Hughes, Joseph D. LaVeigne, and Ivan
Curtis. ”Pothole Detection and Road Condition Assessment using Hy-
perspectral Imagery.” In Proceedings of the American Society for Pho-
togrammetry & Remote Sensing (ASPRS) 2005 Annual Conference, pp.
7-11. 2005.
[6] Jakob Eriksson and Lewis Girod and Bret Hull and Ryan Newton and
Samuel Madden and Hari Balakrishnan ”The Pothole Patrol: Using a
Mobile Sensor Network for Road Surface Monitoring” in The Sixth
Annual International conference on Mobile Systems, Applications and
Services (MobiSys 2008) 2008 June.
[7] Mednis, A.; Strazdins, G.; Zviedris, R.; Kanonirs, G.; Selavo, L., ”Real
time pothole detection using Android smartphones with accelerometers,”
Distributed Computing in Sensor Systems and Workshops (DCOSS),
2011 International Conference on , vol., no., pp.1,6, 27-29 June 2011
doi: 10.1109/DCOSS.2011.5982206