Towards a feature-based photogrammetry

This paper presents strong arguments to use linear and areal features instead of points. In automatic photogrammetry the computer attempts to solve tasks that a human operator performs in traditional and softcopy photogrammetry. Features extracted from imagery are first matched before they enter orientation procedures. The use of features as entities in the orientation is emphasized in this paper. This requires new orientation approaches because the existing methods are point-based. The paper briefly summarizes most recent research in using linear and areal features in orientation. The segmentation and grouping of features is an important prerequisite for object recognition. Fusion of different sensor data also requires the use of features, especially if fusion is performed with information derived from the data.