Using the concept of epipolar geometry in limiting the matching search space to one-dimensional space leads to more efficient matching process especially in the applications including a huge number of points. Resampling of stereo digital imagery into epipolar geometry can be applied after relative orientation so that the rows of pixels in both images lie along epipolar lines. However, original image content might be distorted due to the resampling process.
In this paper, the concept of image epipolarity is employed for matching. But here the images are used directly without resampling them to epipolar geometry. In addition to saving epipolar resampling time, this would have the advantage of preserving the image radiometric characteristics unchanged. The proposed approach is based on finding the epipolar line in one image that corresponds to any specified point on the other image. The epipolar line is searched pixel by pixel, within a search range, for the conjugate point. Experimentation using a stereo pair of digital aerial imagery has shown that matching employing one-dimensional search space and two-dimensional yield slightly different results. The average difference in the y-direction is nearly half a pixel. However, one-dimensional matching results can be enhanced, by following a 3-row matching strategy.
|
