Put another way, if the ball decelerates with timings:
500ms
600ms
700ms
800ms
900ms
1000ms
Then the software knows when to expect the ball to come into view. This way it can better filter out video noise, so it knows what parts of the video are the ball. It is simple for us as humans to do this - to us we just know roughly when the ball will pass, so we know when to look for it. But coding a computer to do it is much more complex.
From tests I did, I did not miss one ball detection at all - not even one. The maximum deviation of timings I encountered was 15ms, which is still well below the 50ms we get from any manual clocking computer. But the average error was about 5ms. So it is about 5 times more accurate than manual clocking roulette computers.
The remaining work before it is available to players:
- refine rotor detection algorithms (not difficult work - 3-5 days)
- wireless transmission module (part that sends prediction to players - fairly straight-forward but tedious testing as we need to ensure it isnt easily detectable, about 1 week)
- software security: to make it just about impossible to use the software without authorization (1 week)
- General bug testing and refinements (1 week)
Rounded off to about a month before full completion/availability. It has all taken longer than expected, as any developmental project does. But we have the wheel tracking part done (finding wheel location and tracking it), and the ball timing algorithms are perfect now. So these are by far the two hardest parts done. I will do a video of these parts for players first chance I get.
I have prepared the hardware so once everything is ready, the package will be complete, and players can arrange to meet for training etc.
