Sport and Augmented Reality
Vision people can enjoy.
When a tennis player challenges a line call, a set of cameras replays the ball's path in 3D, to within a few millimetres. When you try on glasses through your phone, a face mesh tracks you in real time. These are some of the most polished vision systems ever built, and they use the things that you've been learning.
- Line calls to the millimetre
- Trying on glasses by phone
- The most polished systems
How they work
Ball tracking uses many synchronised cameras, and triangulation, the same geometry as our stereo lesson. Player tracking uses detection and re identification. Pitch graphics use homography, drawing on a flat surface. Face filters use keypoints and a 3D mesh. Virtual try on uses segmentation. Every one is a combination of lessons that you've done.
| Application | Built from |
|---|---|
| Line calling | Multi-camera triangulation |
| Player stats | Detection, tracking |
| Pitch graphics | Homography |
| Face filters | Keypoints, mesh |
| Virtual try-on | Segmentation |
Why they feel so solid
Consumer products feel good because of what's hidden. Smooth motion comes from filters that blend each new detection with past ones. High frame rates come from small models. Prediction covers the gap between a measurement and the screen. People notice a wobbly box instantly, so stability often matters more than accuracy.
Summary
Geometry and small models make everyday magic. Next, how did all of this happen? The story of one week in twenty twelve.
Key takeaways- Triangulation tracks balls
- Homography draws on pitches
- Stability is what people feel