Realtime Object Detection Using Ssd

Fatima Ansari
Department of Computer
Engineering
M. H. Saboo Siddik College of
Engineering
Mumbai, India
[email protected]

Mohammed Faiz Khan

Department of Computer
Engineering
M. H. Saboo Siddik College of
Engineering
Mumbai, India
[email protected]
Abstract: Object detection is closely related to
video analysis and image understanding, it has
attracted researchers' attention over the recent
years as technology advanced. Earlier object
detection methods are built on handmade features
and shallow trainable algorithms and architectures.
Their performance easily saturates by giving them
a complex image to computer vision. Human vision
has got millions of years of evolution as a result
human vision is fast reliable and accurate whereas
the computer is century old, with exponential
development of computer into a supercomputer,
there exists rapid progress in deep learning,
powerful tool, which are able to learn semantic
segmentation, high level, deeper features to solve
the traditional problem. Our model uses ssd mobile
net algorithm to accurately detect objects in
images. Our model generates a score in percent of
the object present in the image.

Keywords: object detection, image processing,

realtime object , images

I. INTRODUCTION

Every object detection has two parts Viz Our object detection model uses Ssd mobile net
category recognition and its detection in the algorithm to recognize and detect the object in the
image. Category Recognition deals with image. Our model's algorithm uses appearance
comparing the object present in the image with present in an image to recognize a particular
one of the predefined objects. It is a process in object. The model is relatively simple and easy to
which it will match, compare and identify those
implement. Object detection is a regression
specific objects present in the digital image and
problem that detects directly from bounding box
videos. And category detection deals with
subtracting the object present in the image from its coordinates and class probability. Every object
background. It will enhance the object present in has its own class such as all dish plates are circle
an image. i.e round, which are used in object detection.
In general, object detection algorithms depend
on comparing or matching the object present in the II. LITERATURE SURVEY
image with its predefined dataset. Object
recognition using appearance-based or feature-
based technics. For example, it is used to detect There are many object detection technology came
instances of real-life objects like bottles, into existence in last few years. The main motive
cellphones, etc. In digital images or videos. of object detection is two identify the key features
of the image and leaving out unnecessary things
from an image. Also, there are having many fields
in which detecting the target object and track them
properly is very important. So the need for object
detection increased

A. Survey of the existing system:

There are having many object detection python library opencv. Opencv helps in real-time
algorithms present in the market. Image computer vision. It mainly focuses on image
recognition technology first came into existence in processing, object detection. With the help of
1980. After that many technologies developed in opencv, we can detect 2D ,3D objects easily. With
the image processing field. the help of opencv, we not only identify an object
but also can identify faces and human
Here are some of the technologies of image handwriting.
processing that are previously used

1) ResNet:

Over the last few years, there have been a lot of

discoveries in the field of computer vision. The
computer had achieved the ability of image
recognition and image classification with the
introduction of convnets (Convolutional neural
network) in the 1980s by Yann lecun. For years,
researchers were making the neural network
deeper (adding more layers) to overcome
complex tasks and to improve
classification/recognition accuracy.

But they witnessed, as there were adding more To identify the object properly we will use an
layers to the neural network, it became difficult to SSD(Single Shot multibox Detector) mobile net
train them, and then the accuracy starts to algorithm. SSD mobile net uses a technology in
degrade. Hence to overcome this problem resnet which it divides the object into small boxes and
was introduced. then it will identify the object. With this, we can
identify many objects in a single image. As it can
segment the image into many parts and try to
2) RCNN: identify as many objects as it can. It has more
accuracy than the r-CNN algorithm. It can identify
R-CNN is composed of three modules: any object in real-time. The main benefit of using
Module 1: Region proposal: create and extract SSD mobile net is it can identify images in real-
independent region proposal. time with this we can identify objects present on
Module 2: Feature extractor: get a feature from live images in real-time. Realtime object detection
each candidate region using convnets. is very helpful.
Module 3: Classify features.
The work of R-CNN is to take an image and
identify where the main object (via a bounding A. ALGORITHM
box) is in the image.
1. It takes the image as input from the user via
Problems with R-CNN: webcam.
I. It takes a lot of time as it classifies 2000 2. It processes the image.
regions proposals per image. 3. It extracts all the necessary features from the
II. Cannot be implemented in real-time. image.
. 4. It segments the image in small pieces to
identify more object present in the image.
III. PROPOSED SYSTEM 5. After segmenting it try to identify objects.
6. T starts detecting objects present in the image.
7. It shows the output to the user.
We have used python as a programming language
to code the system. In python, we will use the
B. RESULT

As we run our code it will start a webcam and

start identifying the object which is coming on the
screen. So for example, if a person is sitting in
front of the webcam then the system will detect it Here we can clearly see that it can easily
as a person. Also, it can easily identify many identify more than one object. Also, this is a real-
objects such as a person,cat,dog,bottle,phone, etc. time working model. This will easily identify 3
objects properly. This is a working model.

IV. FUTURE SCOPE

1. This system can used by government in

security cctv cameras in which a security
camera can identify the person who has
any criminal background.
2. Object detection can also be used in
nursery kids learning. It can help kids to
identify the objects.
3. It can also be used in self driving cars to REFERENCES
identify object whle driving a car. It can
help in parking to find a empty parking VI. UNDERSTANDING SSD MULTIBOX —
space. REAL-TIME OBJECT DETECTION IN DEEP
4. It can also be used in iris scan technology. LEARNING
5. This can help many teachers and student HTTPS://TOWARDSDATASCIENCE.COM/UNDERSTAN
for understanding images and various DING-SSD-MULTIBOX-REAL-TIME-OBJECT-
things. DETECTION-IN-DEEP-LEARNING-495EF744FAB
6. It can help for Crowd counting. [1] Introduction to OpenCV-Python Tutorials
7. Object detection can also be used in https://docs.opencv.org/
nursery kids learning. It can help kids to [2] SSD object detection: Single Shot
identify the objects. MultiBox Detector for real-time
processing
[3] Region Based Convolutional Neural
V. CONCLUSION Networks
https://en.wikipedia.org/wiki/Region_Base
d_Convolutional_Neural_Networks
By using the algorithm and OpenCV we are able [4] introduction of ConvNETs (Convolutional
to detect the object present on the screen precisely neural network) in 1980s by Yann LeCun
and able to identify the objects individually with [5] Object Detection Using Convolutional
the proper location pointing to the object. Neural Networks
Different types of objects can be detected and [6] OpenCV Python Tutorial: Computer
identified with the help of the SSD algorithm. Vision With OpenCV In Python
This paper also provides a result that shows the [7] IMAGE RECOGNITION
object detection and identification of objects in APPLICATIONS OF THE FUTURE
real-time.
The real-time object detection is working properly REFERENCE SITES
and giving results also accurate. After getting the
result we can conclude that our project can
identify the real-time objects accurately. [1] https://www.python.org/
[2] www.youtube.com
[3] www.google.com
[4] www,Wikipedia.com
[5] http://researchgate.net/
[6] www.w3schools.com