Partial Object Detection in Inclined Weather Conditions

© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 924
Partial Object Detection in Inclined Weather Conditions
1Margret Sharmila F, 2Amrithya Muralikrishnan, 3Sangavi S, 4Chandrupan R
1Assistant Professor, 2,3,4 UG Student,
Computer Science And Business Systems, Sri Krishna College Of Engineering And Technology Tamil Nadu,
Coimbatore, India.
--------------------------------------------------------------------------***---------------------------------------------------------------------
Abstract- This article provides a comprehensive
analysis of the challenges associated with object
detection. To conduct a systematic examination of
the problem, we present a taxonomy based on the
issues at hand. We delve into each issue in detail and
present a unified view of the solutions proposed in
this document. Moreover, we highlight significant
gaps in the literature that have not been previously
discussed, including existing imbalances and those
that require further exploration.
Keywords: - Imbalance Problems, Object Detection,
Number Plate reading
Object detection is the simultaneous
estimation of class & the location for object instances in
a given image. It is a fundamental problem in computer
vision with many important applications such as
surveillance, autonomous driving, medical decision-
making, and many problems in robotics. Since object
detection is treated as a machine learning problem,
first-generation methods based on manual features and
linear classifiers have maximum profit. The most
successful and representative method of this generation
is the deformable part model (DPM) [13]. The current
generation of OD methods is dependent on deep
learning, and the handcrafted features and linear
classifiers of the first generation methods are replaced
by DL. The above replacement brings significant
performance gains: on the widely used OD benchmark
dataset (PASCAL VOC), DPM achieves a mean average
precision (mAP) of 0.34, while current deep learning-
based OD models achieve around 0.80 mAP. Over the
past five years, although the main driver of OD
advancements has been the incorporation of deep
neural network imbalance problems at multiple levels
in OD, it has also received a great deal of attention. A
given image usually has a small number of positive
samples, but millions of negative samples can be
extracted. If left unaddressed, this imbalance can
greatly affect detection accuracy.
Here two User Personas are addressed, one is
the person who uploads a picture for the Trained
System to detect the Number Plate of the vehicle in
Inclement weather conditions. The other User Persona
is the person who requires the text file that contains all
the Numbers detected.
1. Vision-Assisted Robotic Handling: Object
Positioning, Estimation, Detection, and
Movement Planning
In this article, we will take a closer look at visual
input for robots. We summarize the 3 main jobs of vision-
based robot positioning: object localization, and
perception estimation. Specifically, object localization
tasks include classification-free object localization. Here
the upper operation provides the target object's field in
the input data. The object estimation pose task mostly
refers to 6D object estimation poses. These 3 tasks can be
accomplished by different combinations of robotic
grasping.
2. Modeling contextual scenarios with
Boltzmann machines
Visual modeling is important for robots that
need to see, think, and manipulate objects in their
environment. In this article, we adapt and extend the
Boltzmann machine to simulate a virtual scene. There are
many examples on this topic, but ours is the first to
combine objects, relationships, and space into a powerful
general model. For this purpose, we introduce a hybrid
version of BM. In this model, relationships and offers are
introduced into the model through a common three-way
connection.
3. Object Research using a theoretical model
We delineate an object detection system
based upon a composite model of multiple fault
components. Our system can represent many variable
classes of objects and achieve state results in challenging
PASCAL Object analysis. Although partial decomposition
models has changed into very demanding ones. Our
system is based on a novel approach to discriminative
training using partly labeled data. We combine
techniques for extracting negative data with a method we
call latent SVM.
International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 05 | May 2023 www.irjet.net p-ISSN: 2395-0072
I. INTRODUCTION
II. LITERATURE-REVIEW

4. Imagenet classification using deep neural
networks
We educated a DNN to identify 1.3 million h-res
images in the LSVRC-2010 Image Net sets into 1000
distinct classes. While testing, we attained a top-1 to 5
rate of 39.7% and 18.9%, which are much finer than
previous state-of-the-art results.
5. Benchmark for the 6d proportional object
We offer a model that can estimate a solid's 6D
posture from a single RGB-D input picture. Examples of
3D object mappings or photographs of items, as well as
well-known 6D postures, are included in the training
data. Eight submissions in a single series addressing
major global events, including two new anchor entries,
are included in the prize. at various illumination levels, ii)
a research technique with an error based on blur, and iii)
a thorough review. A constant submission of fresh results
is permitted through an open online assessment system,
which has 15 distinct contemporary modes that reflect
the status of the field. Evaluations reveal that 3D local
feature-based techniques, learning-based methods, and
point-pair feature-based methods are now outperformed
by point-pair feature-based methods in terms of
performance.
6. Finding content through cross-regional
networks
In order to provide accurate and effective
content search, we provide regional integration. Our
regional analysis is remarkably unified, with practically
all comparisons shared throughout the whole picture, in
contrast to the prior regional analysis (e.g. Fast/Faster R-
CNN [7, 19]), which employs hundreds of costly
networks in the region. To do this, we offer unique
features that address the issue of differential translation
vs translation in picture classification. As a result,
frameworks for classifying objects, including residual
networks (ResNets) [10], may be explicitly classified
using our technique. We use a 101-layer ResNet to
provide competitive performance on the PASCAL VOC
dataset (e.g.83.6% mAP on the 2007 set).
Train the system to detect and read the “Number
Plate” of the vehicle and return the Characters detected
in a text file. For further use, the text file containing all
the detected Number plates’ can be used by the requestor
who is a different user persona.
A survey is conducted all over the world to
identify common imbalance problems in objects and
the solution proposed for the corresponding problem is
also recorded. The problem is mainly due to imbalance
occurring in the object during improper weather
conditions, and also due to common imbalance in the
object.
Using GANs rather than duplicating the internet
image is a more significant approach. Example for this
is Task Aware Data Synthesis, the hard sample is
generated using competing networks. Competing
networks include synthesizer, a target network, a
discriminator. Synthesizer is capable of Spoofing the
differentiates, discriminators create high-resolution
digital images to discover the connection. When an
image and ground object is kept, the synthesizer wants
to generate the realistic samples by placing the object
onto the image. To obtain the Realistic image from
synthesizer, discriminator is helped.
Considering the performance of the class during
training, by switching the image of the instance
between the existing images.
Fig 1. Object Detection Process
It is then tested and coded by matching a opaque
concepts sets to boxes of facts. Label at the end Anchor
point is fed to the classification and Regression network
for training. A two-stage approach, First generate object
proposals (or regions of interest) Use anchors through
a separate network
Tools:
Hardware
● System : Pentium IV 2.4 GHz.
● Hard Disk : 40 GB.
● Monitor : 15 inch VGA Color.
● Mouse : Logitech Mouse.
● Ram : 512 MB
● Keyboard : Standard Keyboard
III. PROBLEM STATEMENT

Software
● Operating System : Windows XP.
● Platform :PYTHON
TECHNOLOGY
● Tool : Python 3.6
● Front End : Python anaconda
script
● Back End : Spyder
The object detection literature has made use of a variety
of human-designed metrics and measurements. Metrics
that are learnt directly, however, will give superior
results with intriguing features.
We highlight and suggest numerous significant open
issues and imbalances in object detection in addition to a
thorough discussion of research challenges and solutions.
The imbalances that were studied warrant additional
attention among the many unresolved issues, identifying
new imbalances that have never been addressed or
discussed before.
Fig 2. Open Issues
As emphasized earlier, Foreground-Background Class
Imbalance and Foreground-Foreground Class Imbalance
are the two primary subtypes of the class imbalance
issue, as we stressed before. The issue that has to be
resolved is listed below. Pay closer attention to the view
since it is uneven since fewer people notice it.
Fig 3. Feature-level Imbalance
An excellent example of class inequality. (a)
Model that resembles data (more human models than
parking meters). (b) A case with a distribution that
differs from the dataset. The MS COCO dataset is used for
the images.
More specifically, if we analyze the traditional
FPN The architecture in Figure 9, we note that although
there are several layers of the C2 layer through the
bottom up
Low-level features to the P5 layer of the feature pyramid,
The C2 layer is directly integrated into the P2 layer,
Suggests the influence of high-level and low-level Part
numbers P2 and P5 are different.
Fig 4. Break the balance
(a) Four boxes with negative seals. (b) Two tightly closed
boxes. (c) Indicate how many overlaps there are inside
each pixel of the bounding box. The total number of
samples in the picture Because the pixel is changing, the
frequency of the pixel also changes.
the overlapped amount's bounding box.
Fig 5. SCALE IMBALANCE
Convolutional layers are represented by hierarchical
boxes. A size balancing strategy was not employed. (b)
The prediction is based on Backbone characteristics at
several sizes (e.g. SSD [19]). (c) Before creating
multiscale predictions, intermediate characteristics at
several scales are combined.
IV. RESULT AND DISCUSSION

Fig 6. CAREER LIBRARY
Methods for addressing imbalance issues are categorized
based on difficulties. Please be aware that several
inaccuracies may display in different places if the
currency corrects them all. Virgo R-CNN
In this study, we present a thorough
analysis of the imbalance issue in object detection. We
propose a taxonomy of issues and remedies to solve
them in order to present a more comprehensive and
cohesive picture.
After classifying the issues, we go
through each issue in depth independently and offer
solutions from a united and critical viewpoint. Along with
a thorough analysis of the examined issues and their
resolutions, we also identify and suggest a number of
unresolved issues and imbalances that are crucial for
object detection.
We have found new imbalances that
have not been addressed or discussed before, in addition
to the numerous open parts of the researched imbalance
that need additional consideration.
[1] Q. Fan, L. Brown, and J. Black,"A closer look at fast r-
CNNs for vehicle detection". In 2016 IEEE Symposium on
Intelligent Vehicles.
[2] Z. Fu, Y. Chen, H. Yong, R. Jiang, L. Zhang and X.
Hua,"Backward Optimization and Forward Correlation
Processing for Food Detection", IEEE Transactions on
Optical Image Processing, 2019.
[3] A. Geiger, P. Lenz and R. Urtasun, "Are we ready for
autonomous driving? Kitti Vision Brand, In the IEEE
Conference on Computer Vision and Pattern Recognition
(CVPR), 2012.
[4] X. Dai, “Hybridnet: A fast vehicle detection system for
autonomous driving,” Signal Processing: Image
Communication, vol. 70, pp. 79 – 88, 2019.
[5] P.F. Jaeger, S.A.A. Kohl, S. Bickelhaupt, F. Isensee, T.A.
Kuder, H. Schlemmer and K.H. Maier-Hein,"Retina u-net:
a simple optical application of visual acuity for medical
object detection" arXiv, Volume 1. 1811.08661, 2018.
[6] S.-g. Lee, J. S. Bae, H. Kim, J. H. Kim, "Diagnosis of
Liver Lesions from Underestimation of Large CT Volumes
Using a Single-Tap Multiplexer" Computer Science and
Medical Technology Intervention (MICCAI), A. F. Frangi, J.
A. Schnabel, C. Davatzikos, C. Alberola-López, and G.
Fichtinger, Eds., 2018.
[7] M. Rad and V. Lepetit, “Bb8: A robust, intuitive and
accurate segmentation method for predicting the 3D
representation of complex objects without using depth",
in " IEEE International Conference on Computer Vision
(ICCV), 2017.
[8] W. Kehl, F. Manhardt, F. Tombari, S. Ilic, and N. Navab,
“ssd6d: improved RGB-based 3d analysis and 6d pose
estimation", In the IEEE International Conference on
Computer Vision (ICCV), 2017.
[9] B. Tekin, S. N. Sinha, and P. Fua, “Real-Time Seamless
Single Shot 6D Object Pose Prediction,” in The IEEE
Conference on Computer Vision and Pattern Recognition
(CVPR), 2018.
[10] T. Hodan, F. Michel, E. Brachmann, W. Kehl, A.
GlentBuch, D. Kraft, B. Drost, J. Vidal, S. Ihrke, X. Zabulis,
C. Sahin, F. Manhardt, F. Tombari, T.-K. Kim, J. Matas, and
C. Rother, “Bop: A description for comparing the
positions of 6d" objects, At the European Conference on
Computer Vision (ECCV), 2018.
[11] G. Du, K. Wang, and S. Lian, “Vision-based robotic
grasping from object localization, pose estimation, grasp
detection to motion planning: A review,” arXiv, vol.
1905.06658, 2019.
[12] I. Bozcan and S. Kalkan, “Cosmo: Contextualized
scene modeling with boltzmann machines,” Robotics and
Autonomous Systems, vol. 113, pp. 132–148, 2019.
[13] P. F. Felzenszwalb, R. B. Girshick, D. McAllester, and
D. Ramanan, “Content analysis using machine learning
models", IEEE Transactions on Simulation and Machine
Learning smart, vol. 32, no. 9, pp. 1627–1645, 2010.
[14] A. Krizhevsky, I. Sutskever, and G. E. Hinton,
“Imagenet classification using deep neural networks. In
progress Neural Information Systems (NIPS), 2012.
[15] J. Redmon and A. Farhadi, “YOLO9000: Imagenet
classification using deep neural networks. in progress
Neural Information Systems (CVPR), 2017.
V. CONCLUSION
VI. REFERENCES

[16] R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Rich
character Procedures for accurate object detection and
semantic segmentation" ,In IEEE Conference on
Computer Vision and Models To understand (CVPR),
2014.
[17] R. Girshick, “Fast R-CNN,” in The IEEE International
Conference on Computer Vision (ICCV), 2015.
[18] J. Dai, Y. Li, K. He, and J. Sun, “R-FCN: Object
detection via region-based fully convolutional networks,”
in Advances in Neural Information Processing Systems
(NIPS), 2016.
[19] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. E. Reed,
C. Fu, and A. C. Berg, “SSD: Explore Many One Box", in
Europe computer vision conference (ECCV), 2016.
[20] J. Redmon, S. K. Divvala, R. B. Girshick, and A.
Farhadi,"You Only See It Once: Integrated Real-Time
Performance", in IEEE Congress of Computer Graphics
and Design (CVPR), 2016.

Partial Object Detection in Inclined Weather Conditions

Recommended

Recommended

More Related Content

Similar to Partial Object Detection in Inclined Weather Conditions

Similar to Partial Object Detection in Inclined Weather Conditions (20)

More from IRJET Journal

More from IRJET Journal (20)

Recently uploaded

Recently uploaded (20)

Partial Object Detection in Inclined Weather Conditions