Tennis Vision: Ball Tracking / Player Detection using OpenCV
HTML-код
- Опубликовано: 5 авг 2024
- TRACE is a tool that takes a single tennis match video feed and automatically extracts player, court, and ball information. This information is obtained through a variety of interconnected algorithms that allow us to see the top-view of the tennis match in progress.
github.com/hgupt3/TRACE
I believe your work will bring joy to many people in the future. May your creativity shine bright in the days to come.
Nice work thanks for the video. Can you please answer how can u detect ball bounce
I can see a few frames where it's tracking his balls.
how can i run this code? which file to run?
hi man, can you make a video explaining how you make this programm plis
working on it!
How fast is this algorithm. Can it be used for real-time detection?
If you have a GPU
I test the code with A100. The ball detection with TrackeNet is a little slow. 2 sec/frame. Do you have any idea how to speed it up? Thx@@ffsharsh
What is the version of python?
3.0+ should work
Hi Harsh: Thanks for the video. Can you please answer a few questions so it could help me and other viewers?
1) Did you create the (inset animation) top view ball tracker using only the one video shown here? Is it taken from a single source as in a monocular?
2) Did you perform any camera calibration - soft or hard?
3) Did you have any success getting Z data - ball height in the ball tracking?
4) What is the resolution & FPS of this video
1) Yes, it is monocular.
2) There is no camera calibration.
3) I can only determine when the ball is Z=0. i.e. when there is a bounce.
4) This video is from RUclips, and I did not change it in any way. It was probably 1920x1080 30 FPS.
@@ffsharsh Great. Thanks for the info.
1. Are you able to calculate the X, Y correctly the real world coordinates (not pixels) right...? I am assuming without this you cannot have drawn the plan of trajectory.
2. What is the level of precision? Have you run any ground truths .. to see if they are precise to inches or centimeters?
3. Is it a purely image processing or any network training involved?