IRJET-Road Detection from Satellite Images

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 02 | Feb 2020 www.irjet.net p-ISSN: 2395-0072
© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2497
Road Detection from Satellite Images
Shubham Malpani1, Siddhesh Kamble2, Mangesh Chavan3, Prof. Renuka Nagpure4
1,2,3Dept. of Information Technology, Atharva College of Engineering, Mumbai.
4Professor, Dept. of Information Technology, Atharva college of Engineering, Mumbai.
------------------------------------------------------------------***---------------------------------------------------------------------
Abstract – In this project, we are going to propose a model
of road detection with the help of satellite images. In the
present scenario, roads are detected using Google maps and
navigation systems, but there are some flaws and it is not up
to date and hundred percent accurate. Sometimes roads
detected are in the phase of construction or it is under
maintenance, so the navigation system misleads the person.
It does not tell the person that the provided rote is under
maintenance. So to overcome the inaccuracy and
inconsistency problems, we propose this model of road
detection. In this paper, a method for road extraction and
detection is used. Here we are using Otsu’s method of image
segmentation which is usedoninputimage.Binarization and
Morphological operations are also used. With the help ofthe
above method we can detect roads from satellite images ina
much more efficient manner. The proposed model will differ
between the man-made objects and the roads.
Key Words:Convolutional Neural Networks,Roaddetection,
Satellite Images, Image Processing, Data mining.
1. INTRODUCTION
Roads are one of the most important man-made objects. It is
an important mode of the transport of India. Roads connect
the major portion of the countrytoeachother.Havinga good
transportation and road facilities boost the economy of
country and it helps in transportation of goods from one
state to another or from one city to another . Roads help in
the navigation system.
Roads in India is the second largest network of road in the
world after USA. It is spread over 56, 03,293 kilometer [1].
Roads are partitioned into different sorts, for instance,
Expressways, National Highways, State Highways, District
Roads, and Rural Roads The first evidence of road
development and construction can be traced back in the
2800 BC in the Harappa and Mohenjo-Daro civilization, near
the Indus Civilization .Rural emperors and Monarchs
continued the road construction later when they came in to
rule the country. At present there is allocated budget for
Road and Transportation in the budget of every year.
Expressways are high-speed roads that are four- or more
lanes, and are access controlled where entrance and exit is
controlled by the use of ramps that are incorporated into
the design of the expressway. Most of the existing
expressways in India are toll roads [2]. National Highways
are the roads which connect major city of country to each
other. It is usually denoted by “NH”. State highways are
highways connecting major cities through-out a state. They
also connect roads to National Highways or state highways
of neighboring states. State Highwaysareusuallydenoted by
“SH”. District Roads are the roads which connect the taluka
headquarters and the major roads within a city which are
connected to each other. Rural Roads are the roads which
are in the village and talukas. They are not much developed.
Road helps in the maps and in the online feature of Google
Maps. It helps the user in knowing that which is a shorter
route to travel from a distance A to a distance B. In road
detection, which is basically done with the help of Google
maps and navigation system now a days, it has some errors
in it. Suppose we have to go from a particular place “A” to a
place “B” and there is some construction going on the
desired route or it is under maintenance, the Google Maps
and Navigation System sometimes won’t update that and
even when it is under construction and maintenance it will
tell the user that there is road to go. At present, road maps
are constructed and updated by hand based on high-
resolution aerial imagery but this process is very costly and
time-consuming. Even though we have some error and
mistakes. So, we need a solution for it which shows much
better efficiency in detecting roads. So, in this project we are
going to propose a method to detect roads in a much more
accurate way using Otsu’s method and morphological
operations.
2. LITERATURE REVIEW
2.1 Road Detection from High Satellite Images Using
Neural Network
In this paper, a street discovery model methodology
dependent on neural systems is proposed. The model
depends on Multilayer perceptron (MLP),whichisoneofthe
most favored counterfeit neural system engineering in
arrangement and forecast issues. The RGB esteems are
utilized for choosing the pixel has a place with the street or
not. The discovered street pixels are set apart in the yield
picture. In this paper a methodology of programmed
intersection identification utilizing raster and vector data is
described: mean and standard deviation of dim qualities,
edges as street fringes, and so forth. The determined list of
capabilities was utilized to prepare a feed-forward neural
system, which was the base of the intersection
administrator. The administrator chooses for a running
window about having a street intersection or not.
Neural networks: Neural Networks are made up of simple
processing units called nodes or neurodes [3]. The principle
task related with a neurode is to get contribution from its

neighbors (the yield of different neurodes), register a yield
and send that yield to its neighbors. Neurodes are normally
sorted out into layers with full or arbitrary associations
between progressive layers. There are three sorts of layers:
input, covered up and yield layers responsible for accepting,
preparing and exhibiting the conclusive outcomes
separately.
Multilayer Perceptron (MLP): In this paper, Multilayer
Perceptron (MLP) as a neural network structure is utilized.
One of the most much of the time utilized neural network
models in arrangement is MLP. MLP comprises of a network
of hubs (neurodes) orchestrated in layers. The general
structure of MLP comprises of at least three units organized
in layers of handling hubs: an info layer that gets outer
information sources, at least one concealed layer and a yield
layer that creates the characterization results. Every hub in
MLP can be demonstrated as a counterfeit neuron. In the
MLP, every neuron j in the concealed layer registersthetotal
of information xi weighted by particular association weight
and figures its yield as an element of the whole. In the
preparation procedure, they have utilized 650 haphazardly
rearranged records. MLP is structured in three layers
including an info layer, two concealed layers, and one yield
layer. The information layer has 27 neurons and the
concealed layer has 12 neurons. In the yield layer one
neuron speaking to the street or not. They have utilized the
back-propagation calculation with balanced preparing
parameters (energy and learning rate) as the preparation
strategy and utilized a sigmoid enactment work in all layers
of MLP.
2.2 Road Detection from Remotely Sensed Images Using
Color Features
This work carried out by Automatic detection of roads from
very high-resolution aerial and satellite images is a very
important research field. Unfortunately, the answer is not
straightforward by using basic Image processing and
computer vision algorithms. In this study, they propose a
novel method for automatic detectionofroadsegmentsfrom
very high-resolution color aerial and satellite images. The
method depends on choosing a training set from the input
image manually. It uses color chroma values of pixels as the
discriminative features. Since street pixels have comparable
shading qualities, the circulation of shadingchroma includes
estimations of the preparation area includes a top at a
specific point inside the component space which shows the
road class. Usingthisinformationandone-classclassification
methodology, then they label road segments in a given
remotely sensed image. Finally, they fit a road network
shape on the detected segment.
Detecting the road pixels: In detecting road pixels in this
method, we apply a semi-automatic detection method by
labeling training road pixels for each image separately. In
order to classify road pixels of the input test image, they use
one-class classification method. In classification problems,
generally total number of classes and the label for each class
is known. However, in remotely sensed images, we have an
unknown number of classes such as trees, buildings,parking
lots, various types of roads, agricultural fields, pools and
lakes, etc. If the samples of only one class are known as in
our problem, classificationcannot bedonebya classical two-
class or multi-class classifier, since the other classes are not
represented. This is also the situation for our road detection
problem. Fortunately, a one-class classifier can be used to
separate this small class from other classes. The problem in
one-class classification is to construct a choice boundary to
separate interest class.
Representing Road Pixels: To represent the road network,
they fit lies on to the boundary of detected road segment
pixels. For this purpose, we first extract the Canny edges of
the detected road segments and then discard the edges
which are shorter than 50 pixels. After eliminating short
edges, the road boundary is analyzed to fit the line segment.
The algorithm finds the coordinates of the boundary pixels,
calculates maximum deviation from the line that joins two
junctions or endpoint. If the maximum deviationexceedsthe
allowable tolerance, then the edge is shortened to the point
of maximum deviation and the test is repeated. In this way,
the boundary is represented by line segments.
2.3 Road Network Identification and Extraction in
Satellite Imagery Using Otsu’s Method and Connected
Component Analysis
In this paper, a technique for street identification and
extraction of satellite pictures has been presented. This
technique utilizes the idea of histogram balance, Otsu's
strategy for picture division, associated segment
examination, and morphological activities. The point of this
paper is to find the capability of high-goals satellite pictures
for distinguishing and removing the street arrange in a
vigorous way. Satellite images often contain noise.
Therefore, these images are preprocessed and enhanced
before the extraction of objects. In image enhancement,
digital images are altered and its visual interpretability is
improved. Therefore, resultantimagesaremoreappropriate
for analysis. Contrast enhancement is one of image
enhancement techniques. It improves the appearance of an
object and the brightness between object and its
background.
3. METHODOLOGY
In this paper, a novel method has been proposed for
detecting and extracting the road network from high-
resolution satellite images. In thismethod,theinputimageis
first preprocessed.
Image Preprocessing: - First of all, preprocessing is
performed on the image. Satellite images often contain
noises. Therefore, these images are preprocessed and
enhanced before the extraction of objects. In image
enhancement, digital images are altered and its visual
interpretability is improved. Therefore,resultantimagesare

more appropriate for analysis. Contrast enhancement isone
of image enhancement technique. It improves the
appearance of object and the brightness between object and
its backgrounds.
Grayscale Conversion: - RGB pictures are changed over into
grayscale pictures to decrease cost, handling time and
unpredictability of the pictures.
Histogram equalization: Histogram equalization is a
technique which remaps the input image pixels so that
almost uniform histogram may be achieved. It helps in
enhancing the contrast of the image. It is applied on the gray
scale image.
OTSU method: - Otsu's method for picturedivisionisutilized
for threshold selection. Proposed in 1979. Otsu's method
chooses the limit consequently. It is powerful, quick,
straightforward and stable. In Otsu's calculation, a global
threshold is processed.
Binarization: - After threshold selection, image is converted
into another image where all pixel values which are greater
than the threshold are replaced by 1and restpixel valuesare
replaced by 0. As a result of this step, a black and white
image is formed. Because of this progression, a highcontrast
picture is framed.
Connected component analysis: - Connected pixels areset of
pixels which are not divided by boundary. Its basic idea isan
identification logic whose role is to detect the components
which are single, broken or connected characters. After
finding connected components, trivial openingisperformed.
Trivial opening extracts the connected component based on
some criteria. If connected component of image satisfies the
criteria T, then component is preserved, if not then
component is removed.
Morphological operations: - After connected component
analysis, the removed streets still contain a few openings
and commotions. This is so in light of the fact that associated
segments now and then can't distinguish little ground
objects like structures, paths, vehicles, and so forth. To
dispense with them and improve the exactness of the
outcomes, the removed outcomes are handled by the
different tasks of numerical morphology like opening,
shutting, enlargement, and disintegration. Morphology is a
methodology where items and article highlights are
recognized through their shape.
Fig-3.1: Block Diagram
4. RESULTS
The proposed system was tested by giving input as
satellite images. Following results were acquiredfromthese
tests:
Figure 4.1: Original Image
Original Image is given as input to model. The model works
on this image to give Output.
Figure: 4.2: Grayscale Image

The Original Image is converted into Grayscale form which
reduces parameters to work in image.
Figure 4.3: Histogram Equalized Image
Histogram equalization is used to enhance contrast of gray
scale image.
Figure 4.4: Inverted Image
Histogram equalized image is Color inverted to highlight
edges of Objects in the image.
Figure 4.5: Raw Edge Image
Histogram Quantized image is binarized to produce black
and white image. This image outlines all the objects in the
image.
Figure 4.6: Final output
The final output image is the result All methods used, it
shows proper edges of road. Here we also see shapes and
edges of varios objects and Buildings.
5. CONCLUSION
This project is developed to process the unprocessed high-
resolution satellite images to detect proper highways and
roadways. This model uses algorithms like histogram
equalization and OTSU method to generate desired results.

ACKNOWLEDGEMENT
We would like to acknowledge Prof. Renuka Nagpure, our
mentor, who gave us proper guidance throughout our
project. It is important to acknowledge our guide, as she
helped us in getting proper approach towards or project.
We also would like to acknowledge Dr. S.P. Kallurkar
(Principal), Prof. Deepali Maste (HOD, Information
Technology Engineering), whose encouragement and
guidance played an important role for us in our project.
REFERENCES
[1] "Basic Road Statistics of India 2015-16". Ministry of
Road Transport and Highways. Retrieved 15 January
2018.
[2] "Check out India's 13 super expressways". Rediff. July
2011.
[3] Idris Kahraman, Muhammed Kamil Turan, and Ismail
Rakip Karas, “Road Detection fromHighSatelliteImages
Using Neural Networks,” International Journal of
Modeling and Optimization, Vol. 5, No. 4, August 2015.
[4] Beril Sırmacek and Cem Unsalan, “Road Detection from
Remotely Sensed Images Using Color Features”
[5] P. Yadav and S. Agrawal, “ROAD NETWORK
IDENTIFICATION AND EXTRACTION IN SATELLITE
IMAGERY USING OTSU'S METHOD AND CONNECTED
COMPONENT ANALYSIS,” The International Archivesof
the Photogrammetry, Remote Sensing and Spatial
Information Sciences, Volume XLII-5, 2018.

IRJET-Road Detection from Satellite Images

More Related Content

What's hot (20)

Similar to IRJET-Road Detection from Satellite Images (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET-Road Detection from Satellite Images