High Precision Relative Orientation Using the Feature Based Matching Techniques

This paper presents a method for relative orientation using the feature based matching techniques. The interest operator and feature based matching are use to select points and measure parallax for relative orientation automatically. After the feature based matching, a fine least square matching is preformed to achieve high precision. The matched points are then used for the relative orientation. The preClSlon of relative orientation depends on the final least square matching. An off-line experiment has been carried out. With the optimal window size, and inclusion of radiometric parameters in the least square matching, results of 2.08 ~m for the standard error of unit weight(oo) of the y-parallax for a simulated image model and 3.83 ~m for a real image model are obtained. The estimated precision of the feature based matching together with the least square matching for stereo parallax measurement is 0.98 ~m, or 0.06 pixels. The experiment also shows the high potential for automating the relative orientation procedure.