Exposure Feedback Lets Frame Cameras Estimate Frequencies Far Above the Frame Rate at a Constant Fraction of Photon-Timestamp Efficiency

A frame camera aliases flicker or vibration far above its frame rate; photon-timestamping sensors do not. For cameras with one or two charge buckets per pixel, shutter switches per frame proportional to the highest frequency, and exposure codes adapted to the counts already read, we give a policy that acquires the alias from random strobes, tracks the phase with locked exposure arcs whose number per frame is frozen after acquisition, and ends with an exact-model Poisson likelihood step. We prove that its normalised mean squared error is asymptotically at most a constant times that of photon timestamps, uniformly over flux above a known floor, frequencies in the band and modulation depths in a known interval; with two buckets the guaranteed efficiency tends to 92.3% at high frequency and shallow modulation. A gray-level bisection identifies one of K aliases of a known baseband with O(K(log(1/δ) +1)) photons. On replays of recorded mains flicker above the acquisition threshold, an implementation of the policy, averaged over budgets, stays within a factor of 2 of photon timestamps on the same scoring window in 18 of 24 recording–camera pairs and within 4.6 in all; a non-adaptive strobe baseline is worse in every cell. Exposure feedback thus brings frame cameras within a constant factor of photon counters.