Automated road position alignment of multi-year pavement image sequences with support checks

Objectives Repeated pavement surveys can measure crack growth and treatment effects when images from different years cover the same road section, but positioning and distance differences can shift the match by one or more five-metre images. This paper develops a method that aligns two annual pavement image sequences and reports whether the result supports reslicing.Methods Each image is reduced to a ranked surface-activity value, and a sequence search over the two signals proposes candidate offsets. Each candidate guides local feature matching within twelve-image chunks, and screening retains control points that define a piecewise-linear map reslicing both surveys into matching twenty-metre segments. The first candidate with sufficient control points is accepted.Findings On a U.S. Route 80 archive, automated segment boundaries matched a semi-automatic operator reference to a median of 0.02–0.03 m over the interpolated interior (95th percentile 0.09–0.13 m). After resurfacing, screening retained too few control points. On two further roads, control-point density stayed above the acceptance threshold.Novelty Unlike prior pavement registration that relies on manual landmark selection, the method is fully automatic and makes acceptance explicit, flagging resurfaced or otherwise unalignable pairs for review rather than silently reslicing them.