Why 25 fps is the wrong number to argue about

The argument goes like this: a puck travels at up to 170 km/h, which is 47 metres per second, which at 25 frames per second is almost two metres of travel between frames. Therefore 25 fps cannot track a puck. The arithmetic is right and the conclusion is wrong, because the failure mode is not sampling rate.

Motion blur is the real limit

At 25 fps with a 1/50 shutter, a puck moving at full speed smears across dozens of pixels. The detector does not see a small dark object, it sees a grey streak. Going to 60 fps with the same shutter does not fix that — the streak shortens, but the exposure still integrates the motion. What fixes it is a shorter shutter, which is an arena lighting and camera configuration problem, not a frame-rate checkbox.

Occlusion beats both

In our own error analysis, roughly three quarters of puck-track losses come from occlusion — boards, legs, goalies — and only a small fraction from motion between frames. A faster camera does not help when the puck is behind a shin pad. A second angle and a trajectory model do.

What we actually do

• Track at 25 fps and interpolate between confident observations.
• Fuse multiple camera angles before fitting a trajectory.
• Flag gaps explicitly rather than inventing a smooth path through them.
• Spend the compute budget on pose and re-identification, where it pays.

When frame rate does matter

It matters for the video, not the analytics. Coaches reviewing a stick foul or an offside need 60 fps and a clean shutter, and so do we when a disciplinary case depends on twelve hundredths of a second. That is why the pipeline keeps a 60 fps master and a 25 fps analysis stream, and why they are allowed to disagree about what happened.