The PHODARs - PhotoSynth, Pix4D - Some very cool ideas...

I really like Microsoft's Photosynth product. I have been experimenting with it from the first time I was introduced to at an 4thsight.

From a set of images made in over the course of minutes and submission to a tool like Microsoft's Photosynth images are systematically organized into sets that organize through their 3D relatonships to each other.  Image sets can be purposely taken or they can be disassociated in time, camera, and photographer.  What is especially exciting is that the process also creates a point cloud that can offer quite magical 3D point clouds that can spun and inspected.  And if there are geotagged images in these raw image sets, the entire point-cloud might become geotagged as well as  images that may not have such image geoposition? (There was another form of Photosynth, GeoSynth, that was a bit too powerful for private sector use and resides within a workstation in special places).

The Photosynth above was created from taking 88 images/clips of Uluru (Ayres Rock, Australia) from Google Earth. As you will see all 88 images become properly orientated to the object and each other.   In creating this 3D indexing, a number of camera position orientations are discovered for each image and added to their individual metadata sets. These qualities are valuable to spatial media explorations and include: Image ID, Camera position X, Y, and Z, Camera rotation about X,Y, and Z axis, Image aspect ratio, ans Focal length (relative to the size of the CCD)

The example in the YouTube video was created from grabbing screen from a Gooogle Earth visit to Uluru or Ayres Rock in Australia.  What I was interested  in is the 3D point cloud that can be generated from a set of images... a poor-mans LIDAR solution.  I believe this sort of PHODAR (Photo Light Detections and Ranging) is going to be a significant utility that may render LIDAR solutions as simply too-expensive and too-low resolution. 

The Photosysth Point-Cloud to DEM Tutorial

SynthExport can export this data as a CSV as well as the point cloud. SynthExport allows you to extract the point cloud as well as camera parameter data of a synth on Photosynth. Point clouds are downloaded automatically and converted to formats that are compatible with most 3D graphics applications. Camera information such as position, focal length and lens distortion of each image is stored as a CSV file.   MeshLab and others can load these sets of files and create more dense rendering of the Photosynth magic-data.  And yet another solution VisualSFM that pulls these generalized tools together with a GUI interface.  

The systematization of DEM surfaces from imagery sets  is a hot topic. PhotoModeler is a commercial product that costs between $1,200 to $2,600 The team at 2D3 has something cooking as does Pix4D.  The 2D3 Sensing team is both US and England based and recently Englsnd's DSTL funded 2d3 Sensing to further refine a real-time 3D scene generator to provide rich scene understanding from stereo imagery.  Pix4D is another design supported by Swiss defense interests and is designed to 3D-ize aerial imagery taken from popular, small, and limited range UAVs.  Pix4D is priced on the number of images submitted and the desired resolution of the final 3D model.

One piece of this quest was to discover if these tools might also support and refine the generalized DEM that Google Earth rely on.  What I was able to find is that 3D models generated from these several tools can be converted into a SketchUp form.  SketchUp was once a Google product that was sold to Trimble.  The Trimble version of SketchUp remains compaible to Google Earth; it can absorb Google Earth to its rendings or be sent into Google Earth for that display rendering. Trimble maintains a free version as well as a $495 SketchUp Pro price.

There are several interesting aspects of use of Photosynth derived point clouds is that even though the point cloud that is derived from "your" photos, it is copyrighted / licensed from Micorsoft and can not apparently be used in a commercial manner?  But interesting there may a way around this as the Photosynth Silverlight viewer has a JavaScript API which allows whatever program the synth is embedded in to continually check what the camera angle inside the synth is, so you could match Google Earth's viewpoint to match.

Getting the Best Point Clouds Getting the Best Point Clouds.  There has been debate on the topic.  If you inspect the three comparisons of method - normal (hand-held), Leveled, and All-sorts I sort of fall on the side of Intentionally Crooked.

Stones:
Normal: http://photosynth.net/view.aspx?cid=3034f0ff-0084-45f6-9688-3bc82bf2b958
Leveled: http://photosynth.net/view.aspx?cid=fed01de7-4ad9-432b-8977-ab60f93c521c
Intentionally Crooked: http://photosynth.net/view.aspx?cid=46664b9b-13e6-4f41-a5dc-0256316d7d42

There is an excellent users form that can be found at - http://photosynth.net/forum.aspx. Many of these tools noted above were found from Microsoft by one of the scientists there...

Nathanael Lawrence over 2 years agoThere are actually easier ways to export Photosynth point clouds now.

Almost a year ago, Photosynth user, Christoph Hausner, gave us his SynthExport app.
Kean Walmsley from Autodesk then wrote an import plugin for AutoCAD 2011 called BrowsePhotosynth.

Henri Astre has given us his PhotosynthToolkit, which not only downloads the Photosynth point cloud, but will also download small versions of the photos and camera positions in a synth to use in conjunction with Yasutaka Furukawa's PMVS2 for dense reconstruction of any synth.

Josh Harle has provided a tutorial on how to use PhotosynthToolkit.

If you'd like to use the full resolution photos from a synth (fair warning: you're going to need a heap of RAM to use them in PMVS2 if you have very many), use Henri'sPhotosynthTileDownloader to save them.

Recently, Josh Harle has modified Christoph's SynthExport to export the camera positions to use your photos with 3D Studio Max. His fork of SynthExport is named CameraExport. His results are very similar to the work that Greg Downing from xRez studios demonstrated near the end of November.