A Computer Science comps project by Ethan Ash, Conor Babcock O'Neill, Jack Huffman, Taylor Kang, Doug Pham, and Hannah Scooler. Advised by Professor Eric Alexander at Carleton College. The goal of this program is to track ultimate frisbee players from drone footage and convert it into a 2d animation.
Use the package manager pip to install the following:
pip install yolov5
pip install opencv-contrib-python==4.9.0.80
pip install matplotlib
pip install ffmpeg
pip install PyQt5
pip install pandasPlace an mp4 file name 'frisbee.mp4' in the project directory. Run the following command while in the project directory:
python3 ultimetrics.pyIf you would like to view the tracking process while the program runs (which will take significantly longer to run), then set
test_against_ground_truth = Truthon line 639 of ultimetrics.py before running.
Full readme and directory can be found here https://docs.google.com/document/d/1zZKS5rFvqNjbPdI5Rey-tp2sGcvSVfJuEL2VXnOxmZY/edit?usp=sharing
In order to get the accuracy of the program, set
test_against_ground_truth = Truthon line 641 of ultimetrics.py and then run
python3 ground_truth.pywhere you can annotate the ground truth location of all the players for every 20 frames. You can change the frequency of frames on line 48 in ground_truth.py and on line 642 in ultimetrics.py
Now, when you run
python3 ultimetrics.pyit will print out the average pixel distance away that the tracker is for every ground_truth frame in the video
For questions about the project or contributing to the project, reach out to Eric Alexander.