The app is inspired by the advances in machine learning that can be leveraged for social good. Advances in deep neural networks, inexpensive high quality cameras and high speed cloud computing.
What it does
It assists the visually impaired to navigate their surroundings by telling them about common obstacles like table, chair etc. in their path. It also recognizes people they know personally and notifies them if a friend is in the surroundings.The app detecs the object and an approximation of where the object/person is(left, right). This is done by capturing pictures at an interval of 2 seconds, analysing the image in the cloud and detecting common objects and people in the image. The user can either attach the phone to the cane or wear it around his neck.
How we built it
The product uses an Android App which captures images and communicates to a server. The server is a python module that is hosted on Google Cloud. The python module makes use of google-cloud-vision and open-cv to detect objects and identify people.It uses text-to-speech conversion for notifying the user.
Challenges we ran into
Recognizing people when multiple people are in the frame. Complications while running on Google Cloud(primarily because of inexperience with the platform). Detect
Accomplishments that we're proud of
Recognises tables and chairs with high accuracy and precision. Recognizes people fairly well.
What we learned
Using opencv to detect faces. Running python api on Google Cloud. Running text to speech on phones.
What's next for Open Iris
Right now, the domain of objects that OpenIris identifies is limited to tables and chairs. It can include multiple common objects like garbage can, tree, dog, stairs etc. The people detection accuracy can be improved. Detect dept in images which would enable us to give the number of "steps" to a given obstacle.