Getting Started with Photogrammetry: Photoscan

of 06

Step 1: Getting Ready to Use Agisoft Photoscan for Photogrammetry

In a previous tutorial, we walked through the steps needed to capture photos for use for photogrammetry. This tutorial will use the same set of photos that were used for the previous exercise in order to compare how the two applications differ.
Agisoft Photoscan is an advanced photogrammetry application, which allows for much higher resolution images and larger scenes than 123D Catch. Available in Standard and Pro versions, the standard version is sufficient for interactive media tasks, while the Pro version is designed for authoring GIS content.
While 123D Catch is a very useful tool for creating geometry, Photoscan offers a different workflow, which may be more useful to your project. This is most noticeable in three areas:
Image resolution: 123D Catch converts all images to 3mpix for processing. This offers a good amount of detail in most cases, but may not be detailed enough depending on the scene.
Image count: If covering a large structure or complex object, more than 70 images may be required. Photoscan allows for large numbers of photos, which can be divided up by chunk to balance out the processing load.
Geometric complexity: Photoscan is capable of producing models with millions of polygons. During the processing stage, the model is decimated (programmatic reduction of polygons) down to the number you define.
Obviously these differences come with a cost. First, of course, is monetary. 123D Catch is a free service with premium options for those who require them. Second, the processing power required to calculate the output is all local, instead of cloud-based. To create the most complex models, you may need a multi-processor and/or GPU-augmented computer with up to 256GB of RAM. (Which isn’t possible to install in your average desktop computer… most are limited to 32GB).
Photoscan is also far less intuitive, and requires more knowledge and manual tweaking of settings for optimal output.
For these reasons, you may find it useful to use both tools, depending on what your requirements are. Need something quick & simple, Catch may be a better choice. Want to reconstruct a cathedral with high detail? You may need to use Photoscan.
Let’s get started by loading up Photoscan. (There is a trial available which won’t allow you to save your output if you want to give it a try.)

of 06

Step 2: Load and Prepare the Reference Images

Photoscan’s system, due to its precision, is far less forgiving of skies and other background elements than 123D Catch. While this means more set up time, it allows for significantly more detailed models.
Load your photos into the scene by clicking Add Photos in the Workspace pane to the left.
Use the Shift key to select all the photos, and click Open.
Expand the tree to the left, and you can get a list of Cameras, and indication that they are not yet aligned.
If your photos have any sky visible in particular, or other elements that are not relevant to your model, this is the stage where you remove those elements so that they are not used for processing. This will save you on processing time up front, and cleanup down the road.
Be sure to mask areas where something is in one frame but not another. (For example, a bird flying across the frame in a single shot.) Masking out a detail in a single frame has minimal impact if you have multiple overlapping frames.
Double-click on one of the images, and use the selection tools to select an area, then click “Add Selection”, or Ctrl-Shift-A. Go through all of your images to make sure you have removed unwanted data.

of 06

Step 3: Align the Cameras

Once you have a clean set of camera data, save your scene, close the photo tabs you have opened, and return to the Perspective view.
Click Workflow->Align photos. If you want quick results, choose low precision to start with. Disable pair preselection, and make sure the Constrain features by mask is checked if you masked your photos.
Click OK.
What results is a “point cloud”, which is a series of reference points that will form the basis of your future geometry. Examine the scene, and make sure all the cameras seem to be pointing where they should be. If not, adjust the masking or disable that camera for the time being, and re-align the cameras. Repeat, until the point cloud looks correct.

of 06

Step 4: Preview the Geometry

Use the Resize Region and Rotate Region tools to adjust the bounding box for the geometry. Any points outside of this box will be ignored for calculation.
Click Workflow->Build Geometry.
Choose Arbitrary, Smooth, Lowest, 10000 faces, and click OK.
This should give you a quick idea of what your final output will look like.

of 06

Step 5: Build Final Geometry

If everything looks ok, set the quality to Medium, and 100,000 faces, and recalculate. You will notice a significant increase in processing time, but the resulting detail is well worth the time.
If you have sections of geometry that you don’t want on the final model, use the selection tools to highlight and remove them.

of 06

Step 6: Build the Texture

Once you are satisfied with your geometry, it is time to add the final touch.
Click Workflow->Build Texture.
Choose Generic, Average, Fill Holes, 2048x2048, and Standard (24-bit). Click OK.
When the process completes, the texture will be applied to your model, and ready for use.
In later tutorials, we’ll cover how to use this model in other applications.