Pickleball Referee - Part 1
- 1 minStarted playing Pickleball, I always thought, pickleball was an old person sport, but after the first day of playing close to 2 hours, I had aches and soreness in muscles I did not know existed in my lower half of the body.
The most common problem with the game is everyone arguing over whether the ball was in the court or out.
Enter Pickleball Referee
The idea is to connect 2 Raspberry Pi5 at either ends of the court, attach an AI HAT on top, with a V2 RPi camera, and see if they can act as a referee. I’ve started with CVAT annotations for the video, and converted a 350 frame video to an annotated dataset for YOLO.
train.txt
data/images/train/frame_000000.png
data/images/train/frame_000001.png
data/images/train/frame_000002.png
data/images/train/frame_000003.png
...
...
data/images/train/frame_000350.png
data/images/train/frame_000351.png
data.yaml
names:
0: ball
1: player_1
2: net
3: court
4: kitchen
5: center-line
6: out-of-bounds-line
7: racquet
8: player_2
9: player_3
10: player_4
11: service_area_1
12: service_area_2
13: service_area_3
14: service_area_4
path: .
train: train.txt
labels
frame_000000.txt
0 0.440346 0.643528 0.007526 0.013389 0
1 0.377003 0.686065 0.072901 0.382796 1
8 0.605026 0.612417 0.083594 0.332019 2
9 0.472115 0.463144 0.018073 0.081657 3
10 0.563219 0.481194 0.040646 0.115167 4
frame_000143.txt
0 0.371982 0.462579 0.007526 0.013380 0
1 0.372630 0.618773 0.079396 0.298120 1
8 0.527065 0.553847 0.067703 0.267380 2
9 0.396521 0.471745 0.039052 0.081657 3
10 0.551924 0.466981 0.040641 0.118500 4
After I’ve got the data annotated, I started to look into court mapping and hit upon Homography.
A homography is a geometric transformation that maps points on one flat plane to corresponding points on another flat plane while keeping straight lines straight.
The project is paused for now, need to spend some time finishing the court mapping, before I get back to it. The other challenge is getting the camera with Pi setup and installed at the court!
