I wanted to build a project using an abstracted ML model. Additionally, I felt that this product could actually be used by individuals such as the elderly to stay safe. My neighbor is extremely old for example, and on numerous occasions he would fall in his garage while the car was still on, unable to get up. With a more refined version on FallenAI, the guardian's email could be set to his neighbors and they would be able to help him quickly.

What it does

FallenAI detects when the person being monitored on the camera falls and sends a notification email to the guardian in the case of this emergency. This application would be used to monitor the elderly or the weak, and it has the potential of saving one's life.

How I built it

I built the email API with Golang, the API is called through javascript when the model returns 100% certainly of a fall. I built the neural network model using Tensorflow JS and trained it (comically) by filming myself falling to teach the model what falling looks like.

Challenges I ran into

The main problem I ran into was trying to figure out how to trigger the API call when the ML model was 100% sure of a fall (But I figured it out).

Accomplishments that I'm proud of

Not overfitting or underfitting the model.

What I learned

I learned a lot about TenserflowJS in this project.

What's next for FallenAI

If I was to deploy it, I would probably make it possible to embed the model inside something such as a security camera.

Built With

Share this project: