ElderBot

• 

  Web App

• 

  Telegram Bot

Inspiration

Statistics and past data tell us that around 60% of accidents pertaining to the elderly are not noticed by their guardians for the first few hours. Having been in families with the elderly, we realized a need for a simple but powerful assistant that can alert these guardians.

What it does

ElderBot is essentially an eye that watches over your fortune :). It simply detects if the user in the frame is falling, which, in the unfortunate case if the user is, will notify the associated guardian through a telegram bot. The user may further click the bot's options and call the ambulance or another emergency contact if necessary.

How we built it

The heart of our model is an image classifier that runs real time on the camera stream data. Owing to lack of computational/GPU power, we resorted to leveraging on a pre trained pose-classification dataset by Google. However, we were able to retrain this model with a few images of us(the members) falling/not falling so that the classifier was able to accommodate some edge cases as well. This classifier, along with webapp runs on a Go server (why Go? Because we wanted to try something new!) The notification system runs independently on a Flask server, which uses the Telegram Bot API to send the subscribed user messages on the status of the elderly.

Challenges we ran into

Some of the challenges we faced were during the training of the model, and focusing on how we could reduce the margin for an incorrect classification. We also ran into a lot of API management issues.

Accomplishments that we're proud of

We're definitely proud on what we've achieved for this Hackathon. We believe that this project will definitely serve to a good social use, especially given the simplicity of the idea. We believe that simplicity is what often drives the human mind, something which our project encompasses.

What we learned

One of the most important takeaways was learning how to integrate a ML model with a webapp. This was also our first experience playing around with Tele bots, which is also a huge plus. Additionally, it also allowed us to explore Golang.

What's next for ElderBot

A crucial feature which we considered necessary but was not able to fully integrate was for the Tele bot to send 10 second video snippets before an accident happened. The end product would ideally be running on a standalone camera, which means the application would have to run on a cloud server as well. Another important feature would be to track if the subscribed user has not seen/acknowledged the bot, in which case the bot would automatically send a trigger to the ambulance. We also believe that a variant of this application can be used as Baby monitors!

Built With

• flask
• go
• javascript
• tensorflow