Inspiration
Our daily routine is dynamic. We may be having a slow relaxing walk in the morning, only to find ourselves running to catch the bus a few hours later. So why shouldn't our fitness trackers be as dynamic, giving us more accurate metrics based on the exercise we get by simply living out the day?
What it does
MATCAL utilizes time and speed measurements combined with the power of AI to identify different exercising periods based on intensity, and using all of that to estimate calories burnt during those "workouts".
How we built it
MATLAB of course. The code was split into many smaller functions to help readability and tackle problems one at a time. A test script was used to quickly simulate the function of our tracker, but the different functions can be used and expanded upon depending on future needs.
Challenges we ran into
Figuring out the best Machine Learning Algorithm for "Different Workouts Identification" (that was K-Means with the k being determined by a Silhouette Evaluation). Following up on that, a proper scaling of the data needed to be figured out for better results. (epoch seconds needed to be normalized to be on a comparable scale to m/s speeds)
Accomplishments that we're proud of
Honestly, the fact that the project was completed as a solo entry. Being able to submit a working solution that featured Machine Learning in an innovative way within the given timeframe is more than was initially thought achievable.
What we learned
MATLAB is actually pretty great and intuitive with the use of AI. A serious consideration of adding this to our AI arsenal is worth for the future.
What's next for MATCAL - A MATLAB-based fitness calorie tracker
Right now, MATCAL is only usable by walkers and runners. However a wide variety of workout can be achieved either while standing still, or by doing other exercise (f.e. cycling). In the future MATCAL should grow to correctly identify all forms of workout so that it can include more users.
Log in or sign up for Devpost to join the conversation.