Volume 8, Issue 8, August 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Satellite Image Stitcher using ORB

Viral R. Raval*1, Yash S. Pawar*2, Sohan P. Das*3, Kartik M. Gadagalli*4 Assistant Professor Divya Pritam1,*
*1,2,3,4
Student, Department of Computer Engineering, Shah And Anchor Kutchhi Engineering College, Mumbai,
Maharashtra, India*

Abstract:- Satellite Image Stitcher is a smart system microprocessors that control the camera instead of using a
which will carry out Image stitching and 3D Texture separate processor or PC [3]. The field of computer vision has
Mapping. In this module, overlapping areas in the photos, increased drastically, with the introduction of a variety of
resulting in photographs with accurate information. techniques to perform such tasks [2]. These tasks include
When two images overlap, a portion of the earlier image motion analysis, reconstruction of the scene, image restore,
is duplicated in the present image. If the image cannot be and image matching [4]. This study focused on various image
divided entirely, the degree of overlap on the boundary matching techniques and feature detection algorithms [7]. We
images is increased until they can. The preceding compared how well they performed and how that impacted
module's divided photos with overlapping regions are the stitching process' precision [5]. Some
patched together to create an improved overall image. We algorithms/techniques might work better with certain data
are integrating many photographs with overlapping fields sets, whereas others do not analyze the same datasets as
of vision in this module to create segmented panoramas efficiently. Therefore, some algorithms are useful for a
or high-resolution photos. To achieve accurate results, particular application, whereas others have different
this method of image stitching needs virtually perfect applications in image processing and computer vision
overlaps between photos and identical exposures. This applications.
module sets up the virtual display, creates the sphere on
which the stitched image will be generated, maps the The method of integrating many photographs with
texture images to the sphere, then exports the file as a 3D overlapping fields of view to create an Elongated Image or
viewable HTML file. In this procedure, a three- high-resolution image is known as image stitching or photo
dimensional (3-D) object is wrapped in a two-dimensional stitching. The majority of methods for image stitching, which
(2-D) surface known as a texture map. As a result, the are frequently carried out through the use of computer
surface texture of the 3-D item resembles that of a 2-D software, call for virtually precise overlaps between images
surface.After executing all the phases, we get a 3D and equal exposures in order to achieve seamless results.
Viewable HTML File, where we can view the planet 3 "Stitching" refers to the process of utilizing a computer to
Dimensionally. combine small images into a larger one, ideally without
leaving any trace that the generated image was produced by a
I. INTRODUCTION machine [1].

Nowadays, the development of cameras and multimedia II. MATERIALS & METHODS
technology is accelerating quickly. Also, there is a dramatic
increase in the use of huge images and videos among people. The project's core idea or mode of operation is relatively
Smart cameras used in recent time in a wide range, smart straight forward. The image processing technology is used
cameras are a combination of three parts in one package throughout the project. The project is made up of both static
(cameras, computer vision programs, and image processing). and non-static modules. The system may separate the images
Smart cameras also called intelligent cameras used to hand with overlapping sections, resulting in a higher-resolution
the users for processing the images by providing image. Using texture mapping, a

Fig. 1: Flow Diagram of Proposed System

IJISRT23AUG441 www.ijisrt.com 459

Volume 8, Issue 8, August 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Stitched 2-Dimensional image will be transformed into  A learning method for decorrelating BRIEF features under
a 3-Dimensional texture map model. rotational invariance,leading to better Performance in
nearest-neighbor applications.
This system uses the ORB algorithm.
B. FAST :
A. ORB : A feature detection technique called Features from
ORB builds on the FAST keypoint detector and the accelerated segment test (FAST) is used to extract feature
BRIEF descriptor Algorithm. Both of these algorithms are points from an image. The FAST feature detector has the
attractive because of their good performance and low cost. advantage of being computationally efficient. Superior
performance in terms of computing time and resources can be
ORB’s main contributions are as follows:
attained when machine learning techniques are used. Due to
 The addition of a fast and accurate orientation component its quick performance, the FAST feature detector is ideal for
to FAST real-time video processing applications.
 The efficient computation of oriented BRIEF features
 Analysis of variance and correlation of oriented BRIEF
features

Fig. 2: Features of Binary Vector

C. BRIEF : brief, each keypoint is described by a point vector which is

After detecting keypoint we go on to cipher a descriptor 128 – 512 bits string.
for every one of them. point descriptors render intriguing
information into a series of figures and act as a kind of The Flow Of the ORB Algorithm is as follows :
numerical “ point ” that can be used to separate one point ● The Input Image Is Given.
from another. The defined neighborhood around pixel( ● FAST(Feature From Accelerated Segment Test) Detects
keypoint) is known as a patch, which is a forecourt of some the key points of the image.
pixel range and height. Image patches could be effectively ● Then, The Best points are selected from the key points.
classified on the base of a fairly small number of pairwise ● BRIEF(Binary Robust Independent Elementary Feature)
intensity comparisons. detail convert image patches into a Extracts the Image with patch and then it generates Binary
double point vector so that together they can represent an Feature Vector.
object. A double features vector also known as double point ● After that the final Output Image is Produced.
descriptor is a point vector that only contains 1 and 0. In

Fig. 3: Flow Diagram of ORB Algorithm

IJISRT23AUG441 www.ijisrt.com 460

Volume 8, Issue 8, August 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
III. RESULTS AND DISCUSSION B. 3D SPHERE CREATION & TEXTURE MAPPING
This module sets up the virtual display, creates the sphere
A. Image Stitching on which the stitched image will be generated, maps the
The photos with overlapping regions are patched together texture images to the sphere, then exports the file as a 3D
to create an improved overall image. We are integrating many viewable HTML file. In this procedure, a three-dimensional
photographs with overlapping fields of vision in this module (3-D) object is wrapped in a two-dimensional (2-D) surface
to create segmented panoramas or high-resolution photos. To known as a texture map. As a result, the surface texture of the
achieve accurate results, this method of image stitching needs 3-D item resembles that of a 2-D surface.
virtually perfect overlaps between photos and identical
exposures.

 Heterogeneous Images With Overlapping Parts:

Fig. 4: Images With Overlapping Parts

 The Stitched Image :

Fig. 5: Stitched Image

IJISRT23AUG441 www.ijisrt.com 461

Volume 8, Issue 8, August 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 The Stitched Image Mapped on Sphere:

Fig. 6: Mapped 3-D Image

 Heterogenous Stitched Image :

Fig. 7: Heterogenous Image After Stitching

IJISRT23AUG441 www.ijisrt.com 462

Volume 8, Issue 8, August 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Original Image : Hue And Contrast Image :

Fig. 8: Comparison of Original and Hue & Contrast Image

 Displacement Map :

Fig. 9: Displacement Map of Stitched Image

 Normal Map :

Fig. 10: Normal Map of Stitched Image

C. ORB Algorithm : generates a binary string for each feature based on a

The ORB (Oriented FAST and Rotated BRIEF) algorithm comparison of pixel intensities in a small patch around the
is a feature detection and description algorithm used in image feature.
processing and computer vision.The algorithm combines two
popular methods for feature detection and description: FAST The ORB algorithm improves upon these methods by
(Features from Accelerated Segment Test) and BRIEF introducing orientation information for each feature. This is
(Binary Robust Independent Elementary Features).FAST is a done by computing the intensity centroid of the feature patch
corner detection algorithm that identifies corners in an image and calculating the direction of the gradient at this point. The
by finding pixels whose intensities change rapidly in all orientation is then used to rotate the patch before computing
directions. BRIEF is a feature description algorithm that the feature descriptor using BRIEF.The resulting feature

IJISRT23AUG441 www.ijisrt.com 463

Volume 8, Issue 8, August 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
descriptors are binary strings that are invariant to changes in trilinear filtering to smooth out pixelation or other
illumination and scale, and can be efficiently matched across artifacts.
images using techniques such as the Hamming distance. The
ORB algorithm is also computationally efficient, making it Once the image has been mapped onto the sphere, the
suitable for real-time applications such as object recognition resulting texture can be rendered using a 3D graphics engine
and tracking. or used in other applications such as virtual reality,
augmented reality, or panoramic photography.
D. Why ORB instead of SIFT
Both the ORB and SIFT (Scale-Invariant Feature IV. CONCLUSION
Transform) algorithms are widely used methods for feature
Image stitching is a research area in active
detection and description in image processing and computer
development, with many distinct algorithms for detecting and
vision. However, there are some key differences between the
describing characteristics. This research provides a features-
two that may make ORB a better choice in certain
based image stitching system using the rotation- and scale
situations.One advantage of the ORB algorithm is its
invariant SIFT method, which requires lengthy computation.
computational efficiency. ORB is faster than SIFT because it
The ORB algorithm has a faster execution time and is rotation
uses binary feature descriptors that can be compared using
and scale invariant. In the future, a variety of attributes will
bitwise operations, while SIFT uses floating-point descriptors
be used to provide a thorough inspection of the stitched and
that require more complex computations. This makes ORB
mapped 3-D Model.
more suitable for real-time applications such as object
recognition and tracking.Another advantage of ORB is its REFERENCES
ability to handle occlusion and clutter in images. SIFT
features are scale-invariant, but they are not always robust to [1.] Chung-Ching Lin, Sharath Chandra U. Pankanti,
changes in orientation or occlusion. ORB, on the other hand, Karthikeyan Natesan Ramamurthy, and Aleksandr Y.
computes the orientation of each feature and can handle Aravkin “Natural Adaptive Possible Image Stitching”
partial occlusion by using multiple oriented patches. in IEEE Conference on Computer Vision and Pattern
Recognition (CVPR) 2018.
Finally, ORB is a patented algorithm that is freely [2.] Harrison Chau, Robert Karol “Robust Panoramic
available for use, while SIFT is a patented algorithm that Image Stitching” in International journal of Computer
requires a license for commercial use. This makes ORB more Vision 2019.
accessible for developers and researchers.That being said, [3.] Nakul Shahdadpuri “Real Image of Computer Vision
there are also situations where SIFT may be a better choice, Application and its Impact: Future and Challenges”
such as when dealing with highly textured images or when Wesleyan Journal of Research, Vol 13 No 53
extremely high accuracy is required. December 2020
[4.] Moushumi Zaman Bonny; Mohammad Shorif Uddin
E. 3D Sphere creation & Texture mapping
“Feature-based image stitching algorithms”
Mapping an image onto a sphere is a common technique
International Workshop on Computational
used in computer graphics and computer vision applications.
Intelligence (IWCI) 12-13 Dec. 2016.
This process is also known as texture mapping, and it
[5.] Jaroslav Venjarski; Vivek Dwivedi; Gregor Rozinaj
involves wrapping a 2D image onto a 3D sphere to create a
“Automatic Image Stitching for Stereo Spherical
textured surface.
Image” 2022 International Symposium ELMAR 12-14
Here are the general steps for mapping an image onto a Sept. 2022
sphere: [6.] Wei Zhang; Hao Chen; Wen Chen; Shuting Yang “A
 Define the 3D sphere: First, a 3D sphere needs to be Rooftop-Contour Guided 3D Reconstruction Texture
defined with a certain radius and center. This can be done Mapping Method for Building using Satellite Images”
using 3D modeling software or by programmatically IGARSS 2022 - 2022 IEEE International Geoscience
generating a sphere mesh. and Remote Sensing Symposium 17-22 July 2022.
 Generate UV coordinates: Next, UV coordinates need to [7.] Ashish Mehta, B.K Sriroop, Ramnath Nayak,
be generated for each vertex of the sphere mesh. UV Ravikumar Javali “Virtual Mouse” in International
coordinates are 2D coordinates that represent the position Conference on Circuits, Communication, Control and
of each point on the 3D surface of the sphere. UV Computing 21-22 November 2014
coordinates are typically normalized to a range of 0 to 1.
 Map the Image: The image is then mapped onto the
sphere using the UV coordinates. Each vertex on the
sphere mesh is assigned a corresponding UV coordinate,
which is used to sample the corresponding pixel in the
image. This pixel value is then applied as a texture to the
surface of the sphere.
 Apply texture Filtering: To create a smooth texture on
the sphere surface, texture filtering is often applied to the
image. This can involve techniques such as bilinear or

IJISRT23AUG441 www.ijisrt.com 464