sfm:chapter04

sfm_dense_clean.jpg The Bundle Adjustment process tries to refine and adjust the camera positions and points already processed and add a whole bunch of additional points to the cloud. The Sparse Point Cloud therefore becomes a Dense Point Cloud. Depending on your settings and source material the amount of points can increase multiple times.
A Dense Point Cloud viewed from a normal distance is so dense, that it almost seems solid. If you zoom in however, you still can see individual points. In order to compute a Sparse Point Cloud and the camera positions, we now have to sequentially process the detected features and images. Typically the algorithms start with an “initial pair” where the matches between features are high. This initial pair gets an estimation of the relative positions of the given features. Step by step, more images are added to the algorithm to improve the initial estimation of features or key points. With these images added, the algorithm also tries to solve the individual camera positions through triangulation (Carrivick/Smith/Quincey 2016: 49-50). Once the camera positions have been calculated further processing, can be applied to improve the accuracy of the camera positions before generating a Dense Point Cloud using photogrammetric stereo processing algorithms (Powlesland 2016: 21). A Sparse Point Cloud generated by SfM is often only an intermediary step in the production of much more dense point clouds.

From the Workflow menu select Build Dense Cloud. For this example we will use the Medium Quality settings, to save us some time. Hit the OK-Button.

Additional information: Higher quality settings can be used to obtain more detailed and accurate geometry, but they require longer time for processing. Interpretation of the quality parameters here is similar to that of accuracy settings given in Photo Alignment section. The only difference is that in this case Ultra High quality setting means processing of original photos, while each following step implies preliminary image size downscaling by factor of 4.
The Depth Filtering mode in the Advanced Settings is to find points that somehow where miscalculated, for example due to noisy or blurry photos. If the object has a lot of small details, a mild filtering is recommend. If there are no small details, aggressive filtering is used to save time.

sfm_dense_artefacts.jpgSo if the calculating is done, you should see no difference. That is because there is still the Sparse Cloud showing. In the top menu, find the icon with nine dots, representing the Dense Point Cloud. If you click on that, you will see the difference. A Dense Point Cloud is showing and it should nearly look like the object is already solid. If you have a closer look, you probably will see some free floating chunks of something like in the screenshot on the side here. These are artefacts, that we will need to clean. To do so, we select one of the Selection tools, I usually prefer the Free-Form-Selection. With that, we catch all the chunks that we don't need and delete them with the DEL/ENTF-button on your keyboard. Also make sure to rotate around the object to catch them all by switching to the Navigation tool.

  • sfm/chapter04.txt
  • Last modified: 2024/02/11 19:50
  • by 127.0.0.1