Performance-Based Comparison of Low-Light Video Technologies for Night Surveillance

摘要

Differences in spectral response, pixel size, and noise characteristics between low-light video technologies complicate the direct comparison of field performance for a surveillance application. Commonly measured camera properties such as modulation transfer function and read noise do not directly relate to field performance (e.g., probability of target recognition). In modeling camera field performance based on these measured properties, many assumptions are often made such as constant prereadout gain, signal-independent noise factor, and uniform responsivity. As the imaging task approaches 'photon-starved' conditions, the validity of assumptions made by commercially available modeling software becomes questionable especially for relatively new technologies. Direct comparison of competing technologies can be used to validate modeling results, but such experiments are expensive and time-Consuming, ideally requiring design, fabrication, optimization, and side-by-side testing of complete camera systems under a wide range of potential field conditions.