REAL TIME OBJECT DETECTION USING DEEP
LEARNING
Akula Sagar*, Cherlagudem Tejaswi **, Bhaskari prashanth reddy***,[Link] Kumar Raju
* Information Technology
** [Link] of Engineering and Technology
Abstract- Real-time object detection is a vast, vibrant and technique is used to identify required objects in video
complex area of computer vision. If there is a single object sequences and to cluster pixels of these objects. The
to be detected in an image, it is known as Image detection of an object in video sequence plays a major role
Localization, and if there are multiple objects in an image, in several applications specifically as video surveillance
then it is Object Detection. This detects the semantic applications. Object detection in a video stream can be
objects of a class in digital images and videos. The done by processes like pre-processing, segmentation,
applications of real time object detection include tracking foreground and background extraction, feature extraction.
objects, video surveillance, pedestrian detection, people Humans can easily detect and identify objects present in an
counting, self-driving cars, face detection, ball tracking in image. The human visual system is fast and accurate and
sports and many more. Convolution NeuralNetworks is a can perform complex tasks like identifying multiple objects
representative tool of Deep learning to detect objects using with little conscious thought. With the availability of large
OpenCV(Opensource Computer Vision), which is a library amounts of data, faster GPUs, and better algorithms, we
of programming functions mainly aimed at real-time can now easily train computers to detect and classify
computer vision. One of the crucial components regarding multiple objects within an image with high accuracy.
this is vision, apart from other types of intelligences such Typically is required before touching base at a satisfactory
as learning and cognitive thinking. A robot cannot be too arrangement. This trademark infors that the capacity to
intelligent if it cannot see and adapt to a dynamic plan approaches and rapidly model hopeful arrangements
environment. The searching or recognition process in real by and large assumes a noteworthy part in diminishing the
time scenario is very difficult. So far, no effective solution cost and time required to land at a suitableframework
has been found for this problem. Despite a lot of research execution.
in this area, the methods developed so far are not efficient, Computerized picture preparing is a range portrayed by the
require long training time, are not suitable for real-time requirement for broad test work to build up the practicality
application, and are not scalable to large number of classes. of proposed answers for a given issue. A critical trademark
Object detection is relatively simpler if the machine is hidden the plan of picture preparing frameworks is the
looking for one particular object. However, recognizing all huge level of [Link] are no obvious limits in the
the objects inherently requires the skill to differentiate one continuum from picture handling toward one side to finish
object from the other, though they may be of same type. vision at the other. In any case, one helpful worldview is to
consider three sorts of mechanized procedures in this
continuum: low, mid and abnormal state forms. Low-level
Index Terms- object detection, complex area, computer vision, process includes primitive operations, for example, picture
Localization, semanticobjects, tracking objects, pedestrian preparing to decrease commotion differentiate upgrade and
detection, face detection, ball tracking, Convolution
picture honing. A low- level process is described by the
NeuralNetworks, OpenCV(Opensource Computer Vision).
way that both its sources of info and yields are pictures.
I. INTRODUCTION
Object Detection is the process of finding and recognizing II. RESEARCH AND IDEA
real-world object instances such as car, bike, TV, flowers,
Since the digital image is invisible, it must be prepared for
and humans out of an images or videos. An object
viewing on one or more output device(laser printer,
detection technique lets you understand the details of an
monitor at).The digital image can be optimized for the
image or a video as it allows for the recognition,
application by enhancing the appearance of the structures
localization, and detection of multiple objects within an
within it. Pixel is the smallest element of an image. Each
[Link] detection from a video in video surveillance
pixel correspond to any one value. In an 8-bit gray scale
applications is the major task these days. Object detection
image, the value of the pixel between 0 and [Link] pixel Predicted probabilities are the basis on which these
store a value proportional to the light intensity at that bounding boxes are weighted. A single neural network
particular location. It is indicated in either Pixels per inch predicts bounding boxes and class possibilities directly
or Dots per [Link] resolution can be defined in many from full pictures in one evaluation. Since the full detection
ways. Such as pixel resolution, spatial resolution, temporal pipeline is a single network, it can be optimized end-to-end
resolution, spectral [Link] pixel resolution, the term directly on detection performance. AR technology can
resolution refers to the total number of count of pixels in an address this limitation by providing customers with a more
digital image. For example, If an image has M rows and N interactive and engaging online shopping experience
columns, then its resolution can be defined as MX N.
Higher is the pixel resolution, the higher is the quality of
the [Link] high resolution is not a cost effective
process It is not always possible to achieve high resolution
images with low cost. Hence it is desirable Imaging. In
Super Resolution imaging, with the help of certain methods
and algorithms we can be able to produce high-resolution
images from the the low resolution images.
For example, some retailers have implemented AR
technology in their stores to provide customers with virtual
try-on experiences for clothing and makeup. Customers can
use AR-enabled mirrors or screens to see how clothing or
makeup products would look on them in real-time, without
having to physically try them on. This can save customers Figure1: Types of Image Processing
time and improve their shopping experience.
III. SCOPE OF THE PROJECT
can use. You’ll need to either standardize or normalize
these variables so that they’re within the same range. They
Deep learning is a machine learning technique. It teaches a use many layers of nonlinear processing units for feature
computer to filter inputs through layers to learn how to extraction and transformation. Each successive layer uses
predict and classify information. Observations can be in the the output of the previous layer for its input. What they
form of images, text, or sound. The inspiration for deep learn forms a hierarchy of concepts. In this hierarchy, each
learning is the way that the human brain filters information. level learns to transform its input data into a more and
Its purpose is to mimic how the human brain works to more abstract and composite representation. That means
create some real magic. In the human brain, there are about that for an image, for example, the input might be a matrix
100 billion neurons. Each neuron connects to about of pixels. The first layer might encode the edges and
100,000 of its neighbors. We’re kind of recreating that, but compose the pixels. The next layer might compose an
in a way and at a level that works for machines. In our arrangement of edges. The next layer might encode a nose
brains, a neuron has a body, dendrites, and an axon. The and eyes. The next layer might recognize that the image
signal from one neuron travels down the axon and transfers contains a face, and so [Link] of the synapses gets
to the dendrites of the next neuron. That connection where assigned weights, which are crucial to Artificial Neural
the signal passes is called a synapse. Neurons by Networks (ANNs). Weights are how ANNs learn. By
themselves are kind of useless. But when you have lots of adjusting the weights, the ANN decides to what extent
them, they work together to create some serious magic. signals get passed along. When you’re training your
That’s the idea behind a deep learning algorithm! You get network, you’re deciding how the weights are adjusted.
input from observation and you put your input into one The activation runs through the network until it reaches the
layer. That layer creates an output which in turn becomes output nodes. The output nodes then give us the
the input for the next layer, and so on. This happens over information in a way that we can understand. Your network
and over until your final output signal! The neuron (node) will use a cost function to compare the output and the
gets a signal or signals ( input values), which pass through actual expected output. The model performance is
the neuron. That neuron delivers the output signal. Think evaluated by the cost function. It’s expressed as the
of the input layer as your senses: the things you see, smell, difference between the actual value and the predicted
and feel, for example. These are independent variables for value. There are many different cost functions you can use,
one single observation. This information is broken down you’re looking at what the error you have in your network
into numbers and the bits of binary data that a computer is. You’re working to minimize loss function. (In essence,
the lower the loss function, the closer it is to your desired particular DCNNs, is overfitting, which is poor performance on a
output). The information goes back, and the neural network held-out test set after the network is trained on a small or even
begins to learn with the goal of minimizing the cost large training set. This affects the model’s ability to generalize
function by tweaking the weights. This process is called on unseen data and is a major challenge for DCNNs that can be
assuaged by regularization.
backpropagation. In forward propagation, information is
entered into the input layer and propagates forward through
the network to get our output values. We compare the V. RESULTS
values to our expected results. Next, we calculate the errors
The dense layer (aka fully connected layers), is inspired by the
and propagate the info backward. This allows us to train
way neurons transmit signals through the brain. It takes a large
the network and update the weights. (Backpropagation number of input features and transform features through layers
allows us to adjust all the weights simultaneously.) During connected with trainable weights. These weights are trained by
this process, because of the way the algorithm is structured, forward propagation of training data then backward propagation
you’re able to adjust all of the weights simultaneously. of its errors. Back propagation starts from evaluating the
difference between prediction and true value, and back calculates
IV. THE PROPOSED SYSTEM the weight adjustment needed to every layer before. We can
control the training speed and the complexity of the architecture
A neural network is a network or circuit of neurons, or
by tuning the hyper-parameters, such as learning rate and
in a modern sense, an artificial neural network, network density. As we feed in more data, the network is able to
composed of artificial neurons or nodes. Thus a neural gradually make adjustments until errors are
network is either a biological neural network, made up [Link], the more layers/nodes we add to the
of real biological neurons, or an artificial neural network the better it can pick up signals.
Convolution generates feature maps that represent how pixel
network, for solving artificial intelligence (AI)
values are enhanced, for example, edge and pattern detection. A
problem. The connections of the biological neuron are feature map is created by applying filter 1 across the entire
modeled as weights. A positive weight reflects an image. Other filters are applied one after another creating a set of
excitatory connection, while negative values mean feature maps. Pooling is a dimension reduction technique usually
inhibitory connections. All inputs are modified by a applied after one or several convolutional layers. It is an
important step when building CNNs as adding more
weight and summed. This activity is referred as a
convolutional layers can greatly affect computational time.
linear 25 combination. Finally, an activation function
controls the amplitude of the output. For example, an
acceptable range of output is usually between 0 and 1,
or it could be - 1 and 1. These artificial networks may
be used for predictive modeling, adaptive control and
applications where they can be trained via a dataset.
Self-learning resulting from experience can occur
within networks,
Figure 2: A simple neural network
CNNs and ANN in general use learning algorithms to adjust their
free parameters in order to attain the desired network output. The
most common algorithm used for this purpose is
backpropagation. Backpropagation computes the gradient of an
objective function to determine how to adjust a network’s
parameters in order to minimize errors that affect performance. A
commonly experienced problem with training CNNs, and in
[4] Nikouei, SeyedYahya, et al. "Intelligent Surveillance as an Edge Network
Service: from Harr-Cascade, SVM to a Lightweight CNN." arXiv preprint
arXiv:1805.00331 (2018).
VI. CONCLUSION [5] Thakar, Kartikey, et al. "Implementation and analysis of template matching
for image registration on DevKit- 8500D." Optik-International Journal for
Deep learning based object detection has been a research hotspot Light and Electron Optics 130 (2017): 935-944..
in recent years. This project starts on generic object detection [6] Bradski, Gary, and Adrian [Link] OpenCV: Computer vision
with the OpenCV library." O'Reilly Media, Inc.", 2008. Howard, Andrew
pipelines which provide base architectures for other related tasks. G., et al. "Mobilenets: Efficient convolutional neural networks for mobile
With the help of this the three other common tasks, namely vision applications." arXiv preprint arXiv:1704.04861 (2017).
object detection, face detection and pedestrian detection, can be [7] Kong, Tao, et al. "Ron: Reverse connection with objectness prior networks
accomplished. Authors accomplished this by combing two for object detection." 2017 IEEE Conference on Computer Vision and
things: Object detection with deep learning and OpenCV and Pattern Recognition (CVPR).IEEE, 2017.
Efficient, threaded video streams with OpenCV. The camera [8] Liu, Wei, et al. "Ssd: Single shot multibox detector." European conference
on computer [Link], Cham, 2016.
sensor noise and lightening condition can change the result as it
can create problem in recognizing the object. The end result is a
deep learning- based object detector that can process around 6-8
FPS. AUTHORS
First Author – Akula Sagar, [Link](IT) JBIET and
REFERENCES akulasagar0311@[Link]
[1] Bruckner, Daniel. Ml-o-scope: a diagnostic visualization system for deep Second Author – Cherlagudem Tejaswi , [Link](IT) JBIET and
machine learning pipelines. No. UCB/[Link] UNIV tejaswit034@[Link]
BERKELEY DEPT OF ELECTRICAL ENGINEERING AND
COMPUTER SCIENCES, 2014.
Third Author – Bhaskari prashanth reddy, [Link](IT) JBIET
[2] K Saleh, Imad, Mehdi Ammi, and Samuel Szoniecky, eds. Challenges of
and prashanthreddy194@[Link]
the Internet of Things: Technique, Use, Ethics. John Wiley & Sons, 2018. Internal Guide – [Link] Kumar Raju, [Link](IT),JBIET,
[3] Petrov, Yordan. Improving object detection by exploiting semantic relations and ravisappalaraju@[Link]
between [Link] [Link]ècnica de Catalunya, 2017.