Object Detetcion using SSD-MobileNet

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 06 | Jun 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2668
Object Detetcion using SSD-MobileNet
B.MathuraBai1, Vishnu Priya.Maddali2, Chaithya.Devineni3, Iswarya.Bhukya4 ,
Shirisha.Bandari5
1Associate Professor, Dept.of Information Technology, Vnr Vjiet, Telangana, India
2-5Under Gradute, Dept.of Information Technology, Vnr Vjiet, Telangana, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Object detection is employed in practically every
real-world application, includingautonomoustraversal, visual
systems, and face identification. Real-timeobjectdetectionisa
vast, colourful, and intricate domainofcomputervision. Object
detection is critical in the realm of computer vision. There
have been several improved object detection algorithms
published in literature; nonetheless, the majority of them are
aimed to enhance detection accuracy. As aresult, thenecessity
to reduce computational complexity is frequently overlooked.
These upgraded object detectors require a high-end GPU to
attain real-time performance. In this article, we offer a
lightweight object detection model built on Mobilenet-v2. The
real-time object detector developed here can be used in
embedded systems with limited processing resources. This isa
critical design component of current autonomous driving
assistance systems (ADAS). The suggested lightweight object
detector offers a wide range of applications.
Key Words: Computer Vision, Object Detection,
MobileNet, Single Shot Multi-Box Detector, OpenCv.
1. INTRODUCTION
Object detection is one of the critical regions of studies in
cpmputer vision and prescient today. This is an image
classification technique that aims to detect one or more
types of elements in an image and use a bounding box to
determine their presence.The formatter must build these
components while taking into account the following
requirements. Object detection is the process of identifying
each object in a photograph or image using
computer/software. Object recognition is one of the most
pressing issues in wireless network computer vision. It is
often used in wireless networks and serves as the muse for
complex imaginative and prescient obligations which
includes goal monitoring and scene interpretation. The aim
of object detection is to identify whether there are any
objects in the image that belong to the defined category. If it
exists, the next job is to determine its category and location.
Traditional object detection algorithms are mostly focused
on detecting a few types of objects, such as pedestrian
detection and infrared target detection. Objectidentification
algorithms have made significant progress as a result of
recent advances indeeplearningtechnology,particularlythe
introduction of deep convolution neural network (CNN)
technology. Three primary approaches extensively used in
this sector are You Only Look Once (YOLO), single shot
multi-box detector (SSD), and faster region CNN (F-
RCNN).This study adds to the current literature by
enhancing the accuracy of the SSD method for recognising
tiny items. The SSD algorithm detects largethingseffectively
but is less reliable when recognising tiny objects. As a result,
we alter the SSD method to obtain acceptable detection
accuracy for tiny items. The photos or sceneries were
captured using web cams, and we conducted tests
using common objectsincontext(COCO)datasets.Wecollect
object detection (OD) datasets fromourfacilityforusein our
image processing lab. We create a network using several
libraries including tensorflow-GPU 1.5. Tensorflow
directory, SSD Mobilenet FPN Feature Extractor, object
detection API of TensorFlow, and jupyter notebook areused
for the experimental setup. This overall arrangementallows
us to provide superior real-time object detection. However,
MobileNet with the powerful SSD framework has been a
warm research factor in latest times, because of the
purposeful barriers of running robust neural nets on low-
stop gadgets like mobileular phones/laptops to moreover
amplify the horde of achievable effects with admire to real-
time applications.
2. LITERATURE SURVEY
Celiet. Al presents an object detector based on small sample
learning. The proposed model makes use of object semantic
relevance to improve the accuracy of weak featureobjectsin
complex scenarios. Tanget. Al concentrates on the
framework design and model working principles, as well as
the model's real-time performance and detection accuracy.
Christian szegedy and colleagues provide a simple but
effective formulation of object detection as a regression
problem on object bounding box masks. It describes a
multiscale inference process that, when used by a few
network applications, produces high-resolution object
detections at a cheap cost. Xiaogang wanget offers an
overview of deep learning with an emphasis on applications
in object identification, detection, and segmentation, which
are fundamental difficultiesforcomputervisionwithvarious
applications to photos and videos. A novel object detection
technique is introduced, and objects are further
differentiated by mean shift (MS) segmentation. There is
vision with the assistance of depth information produced
from stereo fixed number of sliding window templates. It is
also possible to use supervised learning to solve the trouble
through manner of approach of imposing it with Decision

trees or, SVM in deep learning, as stated in Malay Shahet.
ZhongQiuZhaoet provides a complete overview of deep
learning-based object identification frameworks that
address various sub-issues such as confusion and low
resolution due to varying degreesofRCNN changes.Sandeep
Kumaret works with the easynet model, which allows for
detecting predictions with a single network. At testing time,
the easynet version examines the complete image, so
predictions are knowledgeable through global context.
3. EXISTING SYSTEM
Prior work was offered to accelerate the spatial pyramid
pooling networks approach. This helped speed up feature
extraction, but it was effectively a forward pass caching
approach. Fast RCNN is a spatially-localized and oriented
method to classify and localize objects in images. It works at
an unprecedented speed, withhighaccuracy,andispowered
by TensorFlow. The model is provided as a single model
rather than a pipeline for immediate training and output of
regions and classifications. The architecture takes an image
as input, processing it to generate a collection of range
recommendations. This collection will be processed by a
deep convolutional neural network with more than 1 billion
parameters and 100 million neurons,inordertogiveyouthe
most up-to-date styles. This procedure will be performed on
top of a pre-trained CNN model. The RoI Pooling layer, close
to the belief of the deep CNN, retrieves traits unique to a
particular input candidate region. The CNN model is first
trained with a pre-trained network, which we provided.The
CNN output is then processed via means of a fully linked
layer, and then the version splits into outputs, one for class
label prediction through a softmax layer and some other
with a linear output for the bounding box. After each region
of interest has been extracted and detected, the next step is
to perform a comparison between eachofthese regions.Fast
RCNN is significantly faster than RCNN in training and test
sessions. When looking at the performance of Fast R-CNN
during testing, incorporating region suggestions greatly
slows down the algorithm. Previously, all object detection
techniques hired areas to discover the object within the
picture. The network does now nolongerstudythecomplete
picture. Instead, regions of the photo with a excessive
probability of containing the object are used. YOLO, or You
Only Look Once, is an object identifying method that differs
notably from the region-primarily based totally algorithms
mentioned above. In YOLO, a single neural network predicts
the bounding containers in addition to their class
probabilities.
4. PROPOSED SYSTEM
The proposed system uses the MobilenetSSDarchitecture to
quickly and efficiently identify objects in real time. A Python
script is written using OpenCV 3.4 that uses a deep neural
network to discover objects. The system works as follows:
The input is real-time video from a camera or webcamwitha
simplified MobileNet architecture that builds a lightweight
deep neural network with depth-separableconvolution.The
input video is split into frames before being sent to the
MobileNet layer. Each feature value is calculated as the
difference between the amount of pixel intensity in the
bright areas and the amount of pixel intensity in the dark
areas. These components are computed using all of the
image's available sizes and areas. Images can contain both
irrelevant and related elements that can be used to identify
items. The task of the MobileNet layer is toconvertthepixels
of the input image into highlightsthatcharacterizetheimage
content. The bounding boxes and related class(label) of
objects are then determined usingtheMobileNet-SSDmodel.
The only remaining step is to display or view the output.
Fig -1: Proposed Methodology Architecture Design
5. RESULTS
This object detection algorithm achieves good results with
any fps low quality camera and can detect objects in real
time with decent accuracy. In our experiment we gave
different images as input and the model has identified then
with a good accuracy. And then we used the webcam to
detect objects in the real-time which also produced the
desired results.
Fig -2: Objects(Person, Car) detected in image

Fig -3: Objects (Animals) detected in image
Fig -4: Objects detected using webcam
Fig -5: Objects detected using webcam
6. CONCLUSION
In this study, we developed a deep learning model for step-
by-step identification of the position of objects in an image.
The framework, like other best-in-class frameworks, could
recognise the item with a good accuracy. In this manner, we
employ an object detection module capable of recognising
what's within the real-time video stream. It uses MobileNet
and SSD frameworks to run modules to provide fast and
productive object detection techniques based on deep
learning. In the future, we can hold to enhance our detection
model , which includes lowering reminiscence use and
growing performance, in additiontoincluding newclassesof
objects.
REFERENCES
[1].Y. Zhong, Y. Yang, X. Zhu, E. Dutkiewicz, Z. Zhou, T. Jiang,
Device-free sensing for personnel detection in a foliage
environment.
[2].S.Z. Su, S.Z. Li, S.Y. Chen, G.R. Cai, Y.D. Wu, A survey on
pedestrian detection.
[3].M. Zeng, J. Li, Z. Peng, The design of top-hat
morphological filter and application to infrared target
detection. Infrared Physics & Technology
[4].L. Deng, D. Yu, Deep learning: methods and
applications.Foundations and Trends in Signal Processing
[5].J. Redmon, S.Divvala, R.Girshick, A. Farhadi,Youonlylook
once: Unified, real-time object detection. In
[6].A. Krizhevsky, I. Sutskever, G.E. Hinton, Imagenet
classification with deep convolutional neural networks. In
[7].Advances in neural information processing systemsH.
Jiang, E. Learned-Miller, Face detection with the faster
[8].R-CNN. In 2017 12th IEEE International Conference on
Automatic Face & Gesture Recognition
[9].R. Girshick, J. Donahue, T. Darrell, J. Malik, Rich feature
hierarchies for accurate object detection and semantic
segmentation. In Proceedings of the IEEE conference on
computer vision and pattern recognition
[10].X. Peng, C. Schmid, Multi-region two-stream R-CNN for
action detection. InEuropeanconferenceoncomputervision
(pp. 744-759). Springer, Cham. (2016, October).
[11].J. Redmon, A.Angelova, Real-time grasp detection using
convolutional neural networks. In 2015 IEEE International
Conference on Robotics and Automation (ICRA)
[12].R. Girshick, Fast r-cnn. In Proceedings of the IEEE
international conference on computer vision

[13].S. Ren, K. He, R. Girshick, J. Sun, Faster r-cnn: Towards
real-time object detection with region proposal networks. In
Advances in neural information processing systems
[14].Y.J. Cao, G.M. Xu, G.C. Shi, Low altitude armored target
detection based on rotation invariant faster R-CNN[J]. Laser
& Optoelectronics Progress

Object Detetcion using SSD-MobileNet

More Related Content

What's hot (20)

Similar to Object Detetcion using SSD-MobileNet (20)

More from IRJET Journal (20)

Recently uploaded (20)

Object Detetcion using SSD-MobileNet