This research presents a proposed strategy of automatic extraction and matching of interest points on stereo aerial imagery. Given a stereo image pair, the strategy starts with applying an interest point operator on both images, followed by suppression of local non-maxima. Each interest point in the master image is then matched against candidate interest points in the slave image. The matching entities are the normalized central moments of orders two and three of interest points. A similarity measure, between the master interest point and each slave candidate point, is estimated based on computed moments. Correct matches are indicated by minimum value of the measure.
A stereo pair of digital aerial images covering an urban area is used for the research experimentation. They are standard aerial photographs that are scanned with a 100 dpi scanning resolution, leading to nearly 254-μm pixel size. The experimentation is carried out on the overlap area exhibited in two 450 pixels by 700 pixels patches of the image pair. A number of 1325 final matches are resulted at applying the presented matching approach. By employing the image coordinates of matched points in a relative orientation process, a global matching precision of 0.6 pixel is yielded.