No Metadata Tutorial - ngageoint/voxel-globe GitHub Wiki

Building a model without any metadata results in model with no scale or location. The SFM algorithm will create camera models centered at latitude and longitude: (0, 0). The model is also in an arbitrary unit-less scale, making it appear much smaller than it really is. Once the model is created, it can be georegistered, which will also set the scale correctly.

Upload

To start a no metadata upload, upload a new dataset and choose "No Metadata" for the Metadata Type. For this tutorial, use the same https://vsi-ri.com/data/downtown_providence.zip providence data, only this time do not upload the cameras and name it "No Meta Providence"

Camera Registration

Once the files have been uploaded and processed, run SFM processing on them. Upon successful SFM processing, the cameras have been registered.

Camera Preview

In order to get a feeling for the cameras, you can use the Image Viewer to look at the cameras.

You will see two sets of cameras in the Tie Point editor now, first the original "No Meta Providence" cameras, which are all at 0,0 pointing straight down.

images/no_meta_wgs84.png

Note In order to see the cameras in Ceisum, you much click the Imagery Icon in the top right corner of Cesium, and scroll down to the bottom and choose the WGS84 Ellipsoid. This will move the surface model so that you can see the cameras now.

Upon clicking on a camera to zoom in on, you will see all the cameras are pointing straight down, and are all on top of each other at the same coordinate, 0,0

images/no_meta_initial_cameras.png

SFM Registration

Now, open up the Image Sets slide out again, and select the "SFM No Meta Providence Image Sequence" Camera Set. Now you can see that the SFM algorithm has positioned and oriented the camera from the no metadata sequence.

images/no_meta_sfm_camera.png

The scale on SFM is quite small (often less than 10 units for an entire scene, which is interpreted as 10 meters), so you will have to zoom in considerably to see that the cameras are registered.

Building a Voxel World

The next step is to start building a voxel world. The bounding box order page will be slightly different. Instead of asking for the bounding box coordinates in latitude and longitude, they are in "units." The bounding box defines the boundaries of the scene containing all the voxels. During SFM processing, matching tie points are stored for every feature matched between multiple images. These tie points are used to determine the location and size of a bounding box including most of the tie points while excluding outliers, such as those circled in red in the image below. This bounding box is stored for use in this Voxel World application.

images/sfm_matches.png

For example, using the "No Meta Providence" data, "SFM No Meta Providence" Cameras, and the "Origin No Meta Providence" Scene, the initial bounding box will be filled in. Roughly +/- 10 in the North/South/East/West direction, +/- 2.3 up and down, and a voxel size of 0.08. These values try to ignore outliers but err on the side of building a large world than typically necessary, but are a good starting point.

Georegistration

Next, create some control points and tie points and run the Camera Georegistration Tutorial.

images/no_meta_geo.png

Now you can run apps like the error analysis on the no metadata data set that you could run only run on full metadata cases.

Building a georegistered Voxel World

Now that the model is registered to the earth, you can build a voxel world like normal.