Artificial Intelligence Based Optical Character Recognition

Using Visual Impaired People Shopping Trolley Technology

K. Geeth Apuroop B. Rakeshwar Reddy T. Chiranjeevi Rao
Department of ECE Department of ECE Department of ECE
SRMIST, Chennai SRMIST, Chennai SRMIST, Chennai
kb3675@srmist.edu.in bb9387@srmist.edu.in tt5367@srmist.edu.in

Abstract Literature Survey

Disability may be a thing where people need to 2.1 An Electromagnetic Sensor Prototype to
depend upon others for his or her works. One help Blind People in Autonomous Walking
among those disabilities is blindness. So far we
We can use electromagnetic sensor to help in
have N number of methods proposed which
walking for blind people. And also helpful for
makes life easy for visually impaired people.
the people who are affected by some visual
Purchasing a product independently is one of the
diseases. It has a radar on the cane making
challenge they are facing on daily basis. To
aware that user is presence in front of the
overcome this problem we came with a solution
obstacle in a wide range and safer range. This
where we use a webcam which captures the
system exhibits noise tolerance and small in
image and using tesseract algorithm text is
size. We used this paper for the survey and
extracted from the image and the text will be
learning of sensor which helps us in distance
converted into an audio which they could hear
calculating. -
through the headphones. On implementing this
technique, during this shopping trolley
technology to spot the merchandise placed 2.2 Reading Aid for the Blind people using
thanks to machine learning and accuracy OCR and Open CV
location find him.
Optical character recognition helps in
Introduction identifying the characters using the camera.
Using this technique blind peoples can Those images are converted to audio output.
move freely future without the assistance of This is the machine translation, text to speech.
other persons, this will be employed AI and computer vision is the field. The
by reading the name of the merchandise and image files will be processed to tesseract and
merchandise details. So that, visual impaired will be converted to text from image and this
people even be ready to shop like normal entire process is using Raspberry pi. This
people. This prototype works during a thanks research is focused on OCR automatic reader
to assist the blind people not only by reading for blinds and people with eye diseases. It
out the but also it calculates the uses python programming as main
space ahead of them. programming language.

2.3 Smart Stick for Blind People with Live

Video Feed
In our life vision plays vital role. It ability
and capability are important. For people who
are visually impaired, they need the help of
others. Instead, the smart system that helps
the blind people to detect obstacle with the
help of blind stick. Raspberry Pi:

2.4 Object Detection Methodologies for blind It's the Microprocessor

people or alittle computer. It consists of RAM,
Processor unit, input ports, output ports,
Vision is that the most vital sense. Image graphics card and also SD
plays crucial role within the human perception card. it's designed to figure on linux also as
of the encompassing environment. Digital on windows platform. It are often operated
image processing is that the field of which it with SSH, or use FTP to transfer the
processes the digital image. the files. it's 1GB RAM. It uses python language
thing identification is that the difficult task for coding. It also provides camera port, USB
for visually impaired people. There are still connections etc.
limitations that need more improving.
It provides the survey and an analysis of OCR:
varied evaluations for the technologies OCR is an application to convert printed
that utilized in the thing identification task. text on the image into machine encoded text
2.5 Text to Speech for the Visually Impaired format.

People who are visually impaired or suffering It is the technique to edit, search, stored the
with eye disease cannot be able to read printed text. OCR uses techniques like as
newspaper or book. So, this system can help “pattern matching”
in reading those by converting text to speech. And “features extraction” to process the printed
Components text on the image captured.

RFID Tags: ESPEAK:

Tags are identified by using It is an algorithm which synthesize

electromagnetic field. There are two sorts speech which works in all platforms like
of tags (i.e. active tag and passive tag) Active windows , Linux and so on . Can change the
tag has its own power supply but the passive sound of the voice by varying characteristics of
tag required for the facility . Passive a text file into pitch range and disturbance to the
tag consists of antenna, chip and means to voice. It uses an ASCII representation. within
gather DC power for reader. In proposed the proposed
system, we've used passive tags system it’s used to convert the document file
because that's less costly and ready to detect obtain because the OCR output into audio file
for area within the super market. for the VIP.

RFID Reader:
RFID reader is employed to Ultrasonic Sensor:
read also as write the info onto the tags. Tag Ultrasonic sensor measure distance by
needed are available the range of reader to using ultrasonic wave and receives the wave
retrieve data from it. reflected back from the thing opposite there
to. It measures the time required during the
emission and reception of the waves to
calculate the space. The device itself work as
emitter and receptor. In proposed
system it's wont to detect the thing or the
obstacles face by blind man.
D. C. Motors:
DC motor contains rotatory electrical motors to
convert DC electrical energy into mechanical Earphone: As an output device
energy. We can control the speed of the motor
Ultrasonic sensor: Detect obstacles
by changing the supply voltage and also by
varying the strength of current in the field.
When we place a conductor, which carries
current in a field containing magnet we can
experience a force. The design of this motor is
far better more when compared to brushed
motors. The complication of transferring energy
RFID Reader: To read the RFID tags
to spinning rotor from outside the motor is
eliminated. It also increases the lifetime and
works with high efficiency without any
maintenance but it contains more risky motor
speed controllers with high cost.

System design:

OCR: To convert image file into text file

Block Diagram:
Module
Raspberry pi: It is used for processing
Image Capturing and Processing
Capturing the image
The respective Webcam can capture an image up
Camera: To capture the image to 30 frames per second. As we are using a USB
type camera, separate driver software’s is not
required and it can be used as play and play. The based on gradient (Sobel – first order
captured image is converted to text and can be derivatives) and two based on Laplacian
heard through a headphone or a speaker which (second order derivative so sensitive to
has 3.5mm jack. Webcam used consists of noise). First, the derivative of image is
features like good resolution, night vision calculated first followed by pointing peak
moreover with a continuous Autofocus. points. Which consists of larger values than
adjacent points the set of extreme points
collectively is an edge. Every image
captured is induced with at least one type of
noise, for increment of efficiency and in
order to reduce the errors the noises are to
Processing the image be filters with suitable filter. For example,
OpenCV offers a wide range of filters for
The image captured introduces noise or poor
reduction of noise. Depending upon nature
quality of page has to be cleared before further
of noise certain filter is chosen to reduce all
processing. This is achieved by processing the
the noises. Edge detection with OpenCV
image. By image processing the pixel density
example is Gaussian blur function.
and quality can be adjusted and corrected.
Appropriate threshold is applied to remove the c) Background Separation
unwanted noise in the image. The number of
pixels are added to objects depend on the size Background separation or subtraction is
and shape of structuring element to process the major step in most of vision-based
image captured. applications. Consider the cases like details
regarding the vehicle is extracted from
Image processing consists of several steps, as traffic camera or a static camera captures the
follows: number of visitors entering the counter etc.
In all above cases necessary step is to extract
a) Filtering
the person or vehicle alone.
It a neighborhood process, in which
Automatic Text Extraction
applying some algorithm to values of pixels
in the adjacent then the pixels of the output The intention of Optical Character Recognition
image corresponding to input is determined. (OCR) is to distinguish optical patterns
This technique helps in modifying or (commonly contained in a virtual photo) with
enhancing an image filter. For example, to respective to alphanumeric or different
remove unwanted features or to emphasize characters. The technique of OCR entails steps
certain features of an image, the only way to like segmentation, characteristic extraction and
achieve those modifications by this process. classification. Considering authentic digital files
in repository, the extraction of signature is
Filtering includes smoothing, sharpening
simple the PDF or PowerPoint shape of
and edge enhancement of the image and
authentic digital files are transformed into
implemented under image processing. suitable high-resolution photo (TIFF, JPEG,
b) Edge Detection etc.) on which signature is computed.

One of the fundamental operations in

performing image processing is edge
detection. This process is useful to reduce
data (pixel) amount to process as well as
structural aspect of image is maintained.
This technique is of two schemes –one
 •        The output of the technique is the binary
photo, there the textual content pixels and
historical past pixels may be visible in
distinctive binary levels.
ü Character Recognition
•        This is ultimate level of the technique this
is person reputation.
•        Binary textual content to ASCII textual
content conversion is being completed here.

Text extraction and reputation technique has 5

steps particularly textual content detection,
textual content localization, textual content
tracking, binarization, and person reputation.
ü Text Detection
•        It takes enter as photo or video component
and test it has textual content or not Tesseract is an open supply textual
content reputation (OCR) Engine. It is well
•        It identifies the textual content having matched with many programming languages and
vicinity withinside the photo frameworks. It analyzes the whole present
format to discover textual content in a large
ü Text Localization document, in any other case it could be utilized
•        Combining the areas containing texts to in conjunction to textual content detector for
assess textual content items and study the spotting textual content from the photo.
bounds across the items is the primary technique
of textual content localization.
ü Text Tracking
•        Only video statistics makes use of the
technique of textual content tracking.
•        It rectifies the output/end result of the
textual content localization and textual content
detection.
•        For the clarity purpose, textual content
embedded withinside the video seems in extra
than thirty consecutive frames.
ü Text Binarization The process of organizing text lines into
blocks, then the lines and the entire region will
•        It is used to split the textual content item
be analyzed for the fixed pitch. In this way word
from historical past in a segmental manner.
finding will be done. Text lines will be done into and obstacle detection are often designed. We
segmental words by the way of character used RFID as well as Raspberry-pi
spacing. It is a two-pass process recognition. technologies to make a smarter environment
Recognizing of each word is made in an attempt for blind people. To unwind the solution to
in the first pass. It will be passed to an adaptive detect the object of interest we’ve used a
classifier as training data when the word is motion based recognition. We use distribution
satisfactory. It recognizes text lower down the of pixel edges and stroke orientation to
page which gets more accurately is the adaptive extract text and Optical Character
classifier. Recognition to recognize text characters
which are converted as audio file and they can
hear through headphones.
Text recognition and output
Off-the-shelf OCR perform
Future Enhancement
text recognition prior to output from
Our future work will extend the text
the localized text regions which are localization algorithm will further more features
informative words. For the an we’ll address the human interface issues
related to text reading by the blind user.
accommodation of characters, the text
region denotes the required rectangular Results

area inside it. The edge boundary of By providing this technology people that
are all suffered from vision impairment may
the required text characters contacts
desire like normal persons and training about
with the border of the text. Proper this device to those people can use this device
margin areas and binary to segment efficiently. This scheme is modernized and
elongated to new form by modernizing software.
text characters are entirely done by the
OCR which generates good
performance. In script files, recognized
text codes will be recorded. Word
recognition is performed by Off the
shelf OCR and converts into audio
output for the visual impaired people.

Conclusion
The proposed system will enable visually
impaired people to buy without others help in Fig.1. image taken by camera
supermarket. Being specific gadget for fig.1.2 Text conversion
product identification, Section information