Edison V2Smart vision for smart people
Indian School of Business
Social Impact Project – 2018/2019
Vadlamudi Vamshi Krishna (71721039), Pathuri Adarsh Reddy (71721010)
Executive Summary
 This project is aimed towards building Smart glasses based on
Artificial Intelligence to help the blind and visually impaired making
their lives easier helping them in various tasks. A personal assistant
built into these smart glasses will assist the user in their day to day
lives based on the commands and queries of the user. The personal
assistant will understand the speech given by the user and provides
the appropriate service or guidance requested by understanding the
context. It alerts the user dynamically by giving them alerts and
notifying the objects or obstacles in their path. It can read books,
newspapers and other documents in front of you just by giving a
command. It can also read stored books and other files whenever the
user needs it.
Business Problem
 According to WHO (World Health Organization), it
is estimated that approximately 1.3 Billion people
suffer from some form of distance or near vision
impairment.
 These people are at a greater risk in experiencing
poverty than other normal people. These people
find it difficult to navigate outside the spaces that
they are used to. They are at a disadvantage
accessing the benefits and services that are
available compared to people without any of the
above problems.
 The aim of this project is to build a solution which
would make their lives easier and independent
enough to lead their lives without having to rely
on others for help in their daily routine.
Solution
 An AI powered smart glasses which serve as a
Personal Assistant.
 Responds to the user based on the commands
and queries asked.
 Understands the context and gives suitable
response to the user.
 Helps in daily commuting and understanding
the environment around the user.
 Reads books, newspapers and other documents
online/offline for the users.
 Plays songs and other audio files based on
command.
Read Documents Listen to Audio Files
Get Dynamic Alerts Scenario Description
Process
Technology
 A lot of image data is needed to train the model for best
accuracy.
 Pre trained models such as Tiny Yolo v3 has been used
here for object detection to make the training process
easy.
 It is the latest version of YOLO which has higher
accuracy than the other versions.
 For image captioning, Flikr8k dataset is used which
contains 8,091 images with captions.
 Yolo v3 is faster and better than other methods such as
RetinaNet, Faster RCNN and SSD.
Architectures
Reading Documents Sensor detection
Architectures
Cloud Scan and read documents
Challenges
 A lot of data is needed to built a perfect
accurate model which takes a lot of time to
train.
 Limited number of datasets available.
 Limited number of objects or classes trained in
the existing datasets.
 Images with captions are limited datasets and
the correctness of these captions is very
important.
 Hardware can be costly depending on camera,
mic and sensors.
 High computational power is needed for
processing so much amount of data.
Why Edison v2?
Edison-V2
Uses a human assistant Uses a AI assistant
Best Solution
mobile App AI assistant
Future scope
 The current project is a prototype which was built with limited amount of
resources and data available. This project can make a huge impact and life
changing for millions of people when the complete product will be developed
in future.
 Advanced features can be built in and it can also serve as a medium through
which the blind, deaf and dumb can communicate with each other.
 The future versions of the product can be used by common man as a smart
device in place of a smart phone when merged with the latest technologies
Edison-V3.
 Collaborating with the government and making sure that this is available to
all the visually impaired could become a revolution in the lives of the
disabled.
Conclusions and Recommendations
 More data can be added to the knowledge database to build a better
model and to increase the robustness of the product.
 Can add more dialogue flows to the already existing ones.
 Keeping note of the user’s input can be used to provide customized
service and better responses.
 Huge processing power and data availability can increase the
efficiency of the product dramatically.
References
 https://m.dailyhunt.in/news/india/english/edexlive-epaper-
edex/madurai+eye+hospital+and+mindtree+foundation+create+special+app+for+children+with+special+
needs-newsid-84205905
 https://www.livestrong.com/article/241936-challenges-that-blind-people-face/
 http://dutchfuturesociety.nl/4466-2/
 https://www.linkedin.com/pulse/robotics-professional-services-revolution-consumer-its-czerniawska/
 https://www.devicedaily.com/pin/how-odgs-smart-glasses-can-help-the-visually-impaired/
 http://www.gizmodo.co.uk/2015/11/microsofts-digital-guide-dog-for-blind-people-sounds-incredible/
 http://edition.cnn.com/2014/06/20/tech/innovation/sonar-sticks-use-ultrasound-blind/index.html
 http://engstema.blogspot.com/
 https://necsus-ejms.org/how-machines-see-the-world-understanding-image-annotation/
 https://medium.com/pyankit/convolution-neural-networks-using-tensorflow-a78d60c1a0ac
 https://sciences.gsfc.nasa.gov/600/highlights/stories/discover.html
 https://aira.io/
 https://www.microsoft.com/en-us/seeing-ai
 https://www.videoblocks.com/video/blind-man-sitting-in-city-park-and-resting-r_u_mkvzbj3p7s5dd
partners
Featured in
⠠⠞⠓⠁⠝⠅
⠽⠕⠥
- Thank you in Braille

EdisonV2 - Smart Vision for Visually Impaired

  • 1.
    Edison V2Smart visionfor smart people Indian School of Business Social Impact Project – 2018/2019 Vadlamudi Vamshi Krishna (71721039), Pathuri Adarsh Reddy (71721010)
  • 2.
    Executive Summary  Thisproject is aimed towards building Smart glasses based on Artificial Intelligence to help the blind and visually impaired making their lives easier helping them in various tasks. A personal assistant built into these smart glasses will assist the user in their day to day lives based on the commands and queries of the user. The personal assistant will understand the speech given by the user and provides the appropriate service or guidance requested by understanding the context. It alerts the user dynamically by giving them alerts and notifying the objects or obstacles in their path. It can read books, newspapers and other documents in front of you just by giving a command. It can also read stored books and other files whenever the user needs it.
  • 3.
    Business Problem  Accordingto WHO (World Health Organization), it is estimated that approximately 1.3 Billion people suffer from some form of distance or near vision impairment.  These people are at a greater risk in experiencing poverty than other normal people. These people find it difficult to navigate outside the spaces that they are used to. They are at a disadvantage accessing the benefits and services that are available compared to people without any of the above problems.  The aim of this project is to build a solution which would make their lives easier and independent enough to lead their lives without having to rely on others for help in their daily routine.
  • 4.
    Solution  An AIpowered smart glasses which serve as a Personal Assistant.  Responds to the user based on the commands and queries asked.  Understands the context and gives suitable response to the user.  Helps in daily commuting and understanding the environment around the user.  Reads books, newspapers and other documents online/offline for the users.  Plays songs and other audio files based on command.
  • 5.
    Read Documents Listento Audio Files Get Dynamic Alerts Scenario Description
  • 6.
  • 7.
    Technology  A lotof image data is needed to train the model for best accuracy.  Pre trained models such as Tiny Yolo v3 has been used here for object detection to make the training process easy.  It is the latest version of YOLO which has higher accuracy than the other versions.  For image captioning, Flikr8k dataset is used which contains 8,091 images with captions.  Yolo v3 is faster and better than other methods such as RetinaNet, Faster RCNN and SSD.
  • 8.
  • 9.
  • 10.
    Challenges  A lotof data is needed to built a perfect accurate model which takes a lot of time to train.  Limited number of datasets available.  Limited number of objects or classes trained in the existing datasets.  Images with captions are limited datasets and the correctness of these captions is very important.  Hardware can be costly depending on camera, mic and sensors.  High computational power is needed for processing so much amount of data.
  • 11.
    Why Edison v2? Edison-V2 Usesa human assistant Uses a AI assistant Best Solution mobile App AI assistant
  • 12.
    Future scope  Thecurrent project is a prototype which was built with limited amount of resources and data available. This project can make a huge impact and life changing for millions of people when the complete product will be developed in future.  Advanced features can be built in and it can also serve as a medium through which the blind, deaf and dumb can communicate with each other.  The future versions of the product can be used by common man as a smart device in place of a smart phone when merged with the latest technologies Edison-V3.  Collaborating with the government and making sure that this is available to all the visually impaired could become a revolution in the lives of the disabled.
  • 13.
    Conclusions and Recommendations More data can be added to the knowledge database to build a better model and to increase the robustness of the product.  Can add more dialogue flows to the already existing ones.  Keeping note of the user’s input can be used to provide customized service and better responses.  Huge processing power and data availability can increase the efficiency of the product dramatically.
  • 14.
    References  https://m.dailyhunt.in/news/india/english/edexlive-epaper- edex/madurai+eye+hospital+and+mindtree+foundation+create+special+app+for+children+with+special+ needs-newsid-84205905  https://www.livestrong.com/article/241936-challenges-that-blind-people-face/ http://dutchfuturesociety.nl/4466-2/  https://www.linkedin.com/pulse/robotics-professional-services-revolution-consumer-its-czerniawska/  https://www.devicedaily.com/pin/how-odgs-smart-glasses-can-help-the-visually-impaired/  http://www.gizmodo.co.uk/2015/11/microsofts-digital-guide-dog-for-blind-people-sounds-incredible/  http://edition.cnn.com/2014/06/20/tech/innovation/sonar-sticks-use-ultrasound-blind/index.html  http://engstema.blogspot.com/  https://necsus-ejms.org/how-machines-see-the-world-understanding-image-annotation/  https://medium.com/pyankit/convolution-neural-networks-using-tensorflow-a78d60c1a0ac  https://sciences.gsfc.nasa.gov/600/highlights/stories/discover.html  https://aira.io/  https://www.microsoft.com/en-us/seeing-ai  https://www.videoblocks.com/video/blind-man-sitting-in-city-park-and-resting-r_u_mkvzbj3p7s5dd
  • 15.
  • 16.