Tennis-Court Simulation in Webots and Isaac Sim
Synthetic tennis-court simulation work in Isaac Sim and Webots for generating ball-tracking datasets and testing camera layouts.
Isaac Sim
- Simulated court scenarios with camera poles, accurate court dimensions, and ball movement animations.
- Built scripts for camera-view simulation, auto-labeling images, and annotation visualization.
- Added camera noise, image saving, and adjustable ball trajectories.
- Targeted challenging cases such as fast balls and distant ball motion.
Webots
- Built a US Open-style court with realistic dimensions and color layout.
- Recorded images and camera poses for tennis ball tracking.
- Solved 3D-to-2D ball projection issues caused by camera-axis differences between Webots and OpenCV.
Dataset Design
| Camera Set | Details |
|---|---|
| Static cameras | 4 pole cameras, 1920x1080, covering both sides of the court |
| Tracking cameras | 2 ball-following pole cameras, 640x640 |
| Train set | Multiple ball trajectories for broader training coverage |
| Test set | Different trajectories for better generalization testing |
Result
- Generated a diverse dataset with over 10,000 images.
- Trained multiple YOLO models on different camera setups.
- Best observed ordering: tracking 2 cams > all 6 cams > static 4 cams.
