We were inspired by a lot of the research going on here at purdue that is machine learning bases and we thought we should give it a try and go for training a model, It also seems like something pretty cool to do and a project we can continue working on in the future.
This project uses Computer Vision and MediaPipe frameworks alongside a trained neural network Keras model that we trained using large amounts of data and machine learning to recognize the up and down of a push-up, detect bad form like butt to high and knees to low, maintains a counter for every correct pushup, and accounts for times when the person in frame is doing nothing by showing "Not in Position" on the screen.
We build it by first researching and keras model that was made for hands and to detect different hand positions, learning from a model we started to make brainstorm how we have to change everything to account for the more nodes and more complication of a whole human body. We first started by writing a python file that is capable of writing all the information we need, the position of all 66 nodes at each capture point and uses that almost 6000 lines of data, to train the Keras model which we had to change because of the higher level of complexity in the body. We than keep improving by adding a counter, and getting more data on different positions we want the model to recognize and we got out final product.
We had an issue integrating the frontend into our project; for some reason, Flask would not cooperate, and no matter what we did, we could not get the Python file to connect to the website. So, we have to run it from the computer file instead, which is practically the same because we still get the computer vision interface and screen that displays all the information.
We were proud that we figured out the Keras model; this is how we trained the program to recognize certain positions, and it required knowledge to understand but creativity to edit and adapt to fit our specific model and specific needs. Training the data to recognize the many positions we had it do was our best accomplishment and the key to our project.
I learned a lot, i never worked with machine learning before this and I have never been more interested in coding, there is so much we can do with it and so much that can come from this overall and is something we am completely excited to work on for future projects and maybe our professions.
We plan to continue working on this project and making it better integrating more data so the model is trained to be more persicince, as well as possibly adding different workouts that the model can detect