The first proper survey of FOD detection in years
A Review of Foreign Object Debris Detection on Airport Runways: Sensors and Algorithms has been published in Remote Sensing. It is the first comprehensive academic survey of this field in a long time, and it is worth the attention of anyone specifying a system.
The value of a review like this is not the novelty — there is none — but the consolidation. Fifteen years of millimetre-wave radar papers, electro-optical work, hybrid architectures and, latterly, deep-learning detectors, assembled into one taxonomy with the trade-offs stated in the open.
Those trade-offs will be familiar to readers here, and it is useful to see them written down by people with no product to sell. Radar sees through weather and does not see low-permittivity objects — which is to say it is excellent at a steel bracket and close to useless at a sheet of bubble wrap. Electro-optical sees anything with contrast and stops working in fog, at night without illumination, and on a wet surface throwing reflections. Hybrid systems exist because neither modality is sufficient, not because fusion is elegant.
The algorithmic half is where the last decade actually happened. Classical detection theory has been substantially displaced by learned detectors, and the survey is properly sceptical about what that has bought: the published results are dominated by the datasets available, and the datasets available are small, clean and unrepresentative of a wet runway at 3am.
What the review cannot supply, because nobody has produced it, is a comparable probability of detection for the fielded systems. Academic work reports detection rates on test sets. Vendors report scan times and object lists. The FAA’s cost-benefit analysis reports detections per month. None of those is the same quantity, and none of them is the one an airport needs to compare two systems.
Recommended reading nonetheless, and a reminder that the research community has kept working on this problem far more consistently than the market has bought it.
Source (MDPI Remote Sensing)