MMS STREET-LEVEL IMAGE DENSE MATCHING

摘要

High density 3D point clouds on the surfaces of buildings and roads can be obtained by dense matching of images taken by cameras installed on a mobile mapping system (MMS) along streets. These point clouds can be used for reconstruction of 3D city models with high level of detail (LoD), and also for mapping of street scenery etc. Plane-based rectification (PBR) uses projective transformation to rectify an image, and is usually used in photogrammetry. Nevertheless, it is not suitable for rectifying image pairs with photographic baseline parallel with the photographing direction. Instead, line-based rectification (LBR) might be adopted for rectifying the afore-mentioned images. This paper trying to improve the points cloud quality which generated from MMS image dense matching, by changing the epipolar rectification algorithm, increasing the redundant observation and integrating multi-image. The experiments are done by using the image dense matching software SURE developed by the Stuttgart University, Germany. Also 24 image sets taken on a MMS are adopted for tests. Their interior and exterior orientation data are determined by photo triangulation. Some tests are done by using three sets of images: 1 .image pairs with photographic baseline perpendicular to the photographing direction, 2. Image pairs with photographic baseline parallel with the photographing direction, 3.Multiple overlapping images including the 1st and 2nd sets. The point clouds determined by using these three sets with different rectification methods will be analyzed. It can be found out that there are some pros and cons among these sets, e.g. 1st set with PBR can get results with higher precision, and LBR can get more information when adding image pairs of the 2nd set, as the rate of successful matching is 43 % and 57% for the 1st and 3rd set, respectively.