A digital twin trained on the physics of foot traffic can spot when someone is feeding a city's pedestrian counters fake numbers - up to a point.
Researchers built a twin that estimates directed pedestrian flows across a street network, folding in a flow conservation law that ties metered segments to unmetered ones. It flags anomalies by combining the twin's prediction error with that conservation residual, and sets its alarm threshold automatically through adaptive conformal calibration rather than a hand-tuned cutoff. Tested against six years of Melbourne pedestrian data, where only 1.18 percent of walkable segments actually carry a sensor, the system cut a single compromised device's worst-case damage to the estimated flow field by 54 percent. Compromise a third of the sensor fleet, and that protection drops to 19 percent.
The stakes are higher than they sound. City pedestrian counts increasingly feed retail and transit planning, safety operations, and economic indicators that get treated as authoritative despite coverage this sparse. Prior stealthy false-data-injection research focused on power grids and water networks, where physical laws pin the system down tightly enough to catch tampering; pedestrian networks are far more rank-deficient, meaning there are many more ways to fake a plausible-looking flow. The researchers show the defense's power comes specifically from the conservation law, not just from knowing which streets sit near which - swap the street graph for a plain distance graph and the protection collapses to 9 percent.
Encouraging, but the math is blunt: a lone bad sensor gets caught more than half the time, a compromised fleet mostly doesn't. Anyone leaning on these counts for something that matters should still ask who controls the hardware, not just trust the twin.