Fidus

PPE Detection for Hospital Environments

Comment

Inspiration

During this time of crisis, it is more important than ever that everyone is staying safe. Seeing the worldwide effects of the pandemic, we wanted to help. We couldn't even begin to imagine the stress that essential workers face daily, having to take every safety precaution and making sure that they don't contribute to the spread of the coronavirus. Sympathizing with their plight, we were inspired by the work of hundreds all over the world making masks and other protective equipment to create an app that would allow them to safely provide professional care to patients in need.

What it does

Fidus serves to take the stress of making sure everyone is wearing the proper PPE off of people's hands and into those of a trained ML model. It takes in an image from a CCTV source, performs a image detection on it, and publishes the results. Used in hospitals or laboratories, Fidus could be useful not only in general but particularly during this pandemic.

How I built it

We used a pre-trained model by ViTech in the aws marketplace. We were originally able to run the sample images through a jupyter notebook running server-side, but wanted to create a more streamlined pipeline. Therefore, we decided to use Amazon Lambda to create a python handler function that is triggered by an API call. This post request handles two of the main parameters, a base64 encoding and the name of the file. A JSON with bounding boxes and prediction probabilities is returned by our lambda function unto the web frontend.

Challenges I ran into

Integrating the API was more difficult than utilizing the image detection model, surprisingly.

Accomplishments that I'm proud of

Due to the challenges we faced, we are proud of connecting the frontend with the backend to create a useful product. We are also proud that we could use AWS SageMaker, a service we didn't know much about prior to Fidus.

What I learned

We learned how to integrate an AWS model into a responsive website successfully and ultimately work together to make a project that can help many people around the world.

What's next for Fidus

Taking in video input, sending 10 second intervals to the lambda function and allowing us to handle real CCTV footage. Better web UI. Alert notification system using Twillio or something similar.

Share this project:

Updates