Sensitivity analysis for correlation-based registration of digital images

Traffic management systems, which derive their measurements of vehicle counts, class, speed, and occupancy from TV sensors, are growing in popularity. The advantage of such systems is that, as new lanes are defined, the information of detection zones can be reprogrammed without having to physically move any embedded sensors. However, such TV cameras are frequently used in the dual roles of a traffic monitor, and of a traffic surveillance sensor. In the latter role, the TV camera is often moved in some pan or tilt, and even zoom mode. Following such a use, the camera is returned to its traffic monitor role. For many systems, this return to the traffic monitor mode requires a manual realignment. For other systems, this alignment is automatic and based upon a straightforward use of correlation or a variant of a modified correlation scheme. The question the system designer is required to answer is that of how accurate is this realignment. The answer is a function of measurable image variables. What is presented below is the theoretical analysis of that correlation process. This analysis allows the designer to tradeoff the selection of the correlation process such that accuracy might be maintained in spite of changes in such variables as signal-to-background measurements, signal-to-noise, and average pixel-to-pixel correlation.