Inspiration
We started off just looking at the hardware we had, and trying to think of a way to produce something that could be genuinely advantageous with what we had available. The proximity sensor we had on one of our boards seemed interesting, and we realised it could be valuable for accurately tracking hand tremors.
What it does
TrembleTrack modules are small, cheap (our prototype uses a Raspberry Pi and a >£10 microcontroller board) interfaces that can be placed in the rooms of care receivers at a care home. A simple interface using coloured LEDs allows patients to provide daily data on hand tremors by holding their finger over a proximity sensor for a few seconds. This data is then processed and passed through a simple logistic regression machine-learning algorithm to classify it by level of concern, and uploaded to a centralised server, where our intuitive frontend interface can be used to track trends in the data and catch risk early. The low-cost aspect of the modules is particularly important - the system should be simple and cost-efficient to set up (especially in bulk), making it a perfect project for philanthropy or sudden funding.
How we built it
The TrembleTrack modules consist of a Raspberry Pi connected to an Adafruit Proximity Trinkey microcontroller board. When activated, the proximity sensor on this board will track finger proximity for several seconds, and then send the data to the Pi which will calculate the standard deviation and total distance moved by the finger. We then use logistic regression from scikit-learn to classify the data based on concern level, and send this data on to a central server designed in Go. The web frontend queries this server to display the data, concern level and a graph over time to carers.
Challenges we ran into
Machine learning is pretty new to us! While scikit-learn provides a very nice wrapper, it's outside all our areas of expertise so it still took a while to get things running.
We also had some issues with certificates hooking up our frontend and and backend due to overeager web security.
Accomplishments that we're proud of
None of the team members that worked with Go and Flutter had ever used them before, and we're very happy with the code we produced! We also didn't have that much hardware with us, and we're excited that we've made something that we genuinely think has the potential to be useful from what we had.
What we learned
We learnt a lot about classifiers, and what's going on behind the scene that makes different models better for different tasks. We also learnt a lot of Go! With Flutter, we learnt how to build layouts, various data visualisation tools, and dealt with HTTP querying.
What's next for TrembleTrack
Carehome domination! Or maybe just getting a big dataset for tracking trembles - by taking some samples from people with varying conditions we could probably significantly improve the accuracy of our model. Adding age as another feature would probably help even more, so that we aren't concerned about harmless tremors in the elderly. We'd also like to add better infrastructure for multiple users.
Log in or sign up for Devpost to join the conversation.