What if i told you that every month, 1 billion users use Google Maps to venture into new places independently, yet 33 million are left baffled. Why? Because of lack of accurate accessibility of the app for the differently abled population- the blind and deaf. Our vision is equal accessibility to navigation!
What it does
DigiCane is an Android application that make navigation accessible to the blind and deaf by combining obstacle detection with navigation. Our extensive customer research highlighted two main problems - traffic signal detection, audible information overload with lack of haptic responses. Our app solves these problems by hiding information about obstacles that are not in proximity and including vibrations as responses.
How we built it
We trained Tensorflow's Single Shot Detection model for realtime obstacle and traffic light detection on Android. The model was trained on MS COCO dataset and manually curated images of road obstacles.
Challenges we ran into
Customer Discovery, optimal obstacle detection, traffic sign detection, training deep models with extensive data, data collection and cleaning
Accomplishments that we're proud of
Took close to 10 customer interviews for deep insight about the problem definition, Extremely low latency, good response from our customers and a successful prototype makes our dream of social good by bringing accessible navigation and independence to the sightless, a reality.
What we learned
Customer needs and interface, customer research, better models, elimination of information which is not useful and need for interaction with users sing both vibrations and voice
What's next for DigiCane
Improving accuracy of model and testing with real users for more pivots and improvements