TrickTracker

• 

  object detection

Inspiration

We wanted to do something machine learning related and we all have a passion for skateboarding so we put our skills to use!

What it does

TrickTracker is a webapp that classifies and outputs information about people’s skate tricks based on video input from a webcam. It can accurately tell the difference between an ollie, kickflip and a shuvit. It can even tell how high the user jumped when on the board!

How we built it

We used a Python backend with a React JS frontend to render our app. As for our machine Learning components, we used Tensorflow to create the video classification model for skate tricks and YoloV8 to create our object classification model and track the location of the skateboard.

Challenges we ran into

We were trying to do real-time object tracking with OpenCV and a raspberry pi but we found out that it was very tedious to do, especially on a brand new device. Additionally, it had slower processing than a regular computer so we had to stop midway and continue with just our computers.

Accomplishments that we're proud of

We successfully implemented object tracking and height tracking in a clever way in our program.

What we learned

To think ahead about the technologies you are going to use in your project.

What's next for TrickTracker

We are going to extend the program to make it so that it can more accurately detect height, detect a variety of more tricks, as well as detect with an unstable frame.

Built With

• javascript
• opencv
• python
• react
• tensorflow
• yolov8