Fire a projectile into a target at hypervelocity and it throws off a cloud of fragments. High-speed cameras capture the event, and what aerospace engineers need from that footage is the distribution of debris mass and speed, because that is what determines whether a shield protects a spacecraft.

The imagery is unforgiving. Many fragments, all moving extremely fast, in noisy footage from a highly specific instrument, so the general-purpose tracking methods developed on everyday video have little to work with.

What makes this a useful application paper is that the authors document how they extended an off-the-shelf tracker rather than presenting a bespoke system. That is the more transferable contribution: the next team facing a similarly odd domain learns which adaptations were needed and why, instead of receiving a black box tuned to one experiment.