An iOS app for empowering visually impaired to help to find their belongings and overcome day-to-day struggles through the Mobile camera as "Third Eye" using Object Detection Algorithm and Speech Recognition.
One morning my grandma woke up and could not find her glasses and medications although they were just on the other side of the bedroom; I once met a visually impaired couple in Perkins Hackathon who shared their story of day-to-day struggles in their life finding items in their house when they forget to keep each items in the right place/ location. We made this app in the hopes to accommodate people like my own grandma and the millions of visually impaired people around the globe.
Our project takes input from User to find a specific item using speech-to-text and detects the requested object in the room in "REAL-TIME" and using beep sensing and vibration in the mobile phone it will guide the user to the item.
Using Object Detection algorithm via Pre-trained RESNET50 Deep Learning model and inbuilt iOS Speech-to-Text Feature and incorporating multi-threading to perform multiple activities at the same time
- Detecting object in "real-time"
- In order to decrease the latency of response from the deep learning model, we had to integrate the neural network with the iOS platform and deploy the model on the mobile phone itself, instead of relying on a cloud service and getting detection results from the cloud
- Choosing the most-fit Object Detection Algorithm
- Integrating the neural network on an iOS platform instead of on a cloud service
Detecting objects in realtime and less response time with the Machine Learning model deployed itself on the iPhone. We are proud of what we vision, we were able to put in reality and can help our target audience now.
We learned about Machine Learning apps that support development on iOS along with build-in features and various Object Detection algorithms and their compatibility with the iOS app.
Extending our app using Drone which can help to find items without any human support and can freely rotate in 360 angles. It would be interesting to use a combination of MobileNet and SSD to detect an object in real-time video streaming.
Xcode 11.2 | Swift 5.1 | Python | AVFoundation | Vision | Speech | UIKit