I don't have high speed camera but clay courts
and it's better for doing accuracy test.
I made such tests with the 2018-02-02 firmware and with the latest 2018-05-05.
And there is great improvements between them.
With the 2018-05-05 firmware, the average difference, in cm, between the position computed from the device and real position is
6.6cm in the length of the court and
3.8cm in the width. Even better if the bounce is in the half court near the device :
4.5cm in the length and
2.2cm in the width (against
15cm and
18.3cm with 2018-02-02 firmware).
The results are detailed in the two following pictures :
- test with 2018-05-05 firmware
- test with 2018-02-02 firmware
And my process is detailed in the following picture:
The accuracy is starting to be very interesting. And don't forget that there is another process for linecall : pure computer vision to "see" if the ball is "before" or "after" the line.