Point Clouds from Multiple Sensors: The View Point Inn, Historic Preservation

The View Point Inn, located in Corbett, Oregon, is listed on the National Register of Historic Places. In 2008, the inn was used in filming of prom scenes in the first film of the Twilight Saga film series. The exposure generated by the film was credited with giving the hotel and its restaurant a boost in business.

In July 2011, the building's second floor was gutted by a fire, and its owners, Geoff Thompson and Angelo Simione, faced financial difficulties, as the insurance on the building had lapsed. In October 2013, plans were made to rebuild and reopen the historic structure, with work slated to begin the winter of 2013–2014. Restore Oregon listed the inn on its 2012 list of Endangered Places; in December 2014 it reported that restoration was unlikely. (Wiki)

Recently however, Peter Meijer Architect, PC (PMA) contracted ToPa 3D to begin the initial 3D building survey of this historic site.

After-fire photo from Matterport (360o Panorama Mode) - by ToPa 3D

The assessment process has begun and it looks like things are underway to restore this property. Having had lunch there myself before the burn, this author can attest to it's classy beauty and prime location in the Columbia Gorge.

The first order of business was to create as-builts for the structure with 3D scanning since these records did not exist. For data collection leading to the as-built drawings, ToPa 3D used 3 separate "reality capture" sensors to create point clouds (3D point models): A FARO Focus3D scanner, the Matterport 3D Showcase scanner, and a DJI Inspire Pro Drone. The FARO scanner was used to document both the interior and exterior conditions requiring high accuracy. The Matterport scanner was used to create a virtual tour to provide the project 3D modelers at PMA with numerous vantage points and to understand the interior space context quickly. Finally, the Inspire was used to survey the entirety of the project site to the borders of the property for a total site plan.

The total field time to accomplish all of these field tasks was about 14 hours... That's 76 color FARO Scans, Matterport Virtual Tour, and drone site mapping.

Matterport Documentation

Visit the View Point Inn Matteport 3D Showcase Here

FARO Documentation

Using the FARO Focus3D laser scanner, ToPa 3D scanned both the interior and exterior areas of the View Point Inn's property. PMA would use this to accurately identify the location of the building and fountain areas as they relate to one another, and create the as-builts from the point cloud.

FARO point cloud looking east

FARO point cloud looking west

ToPa 3D scanned this entire property without targets and relied on top-down and cloud-to-cloud registration. We did find that this method performed very well in the interior spaces where there was a higher density of scans. For areas with low overlap, the algorithms struggled a bit, requiring us to manually pick a few planes and points to bring clusters together.

With the images presented in this blog, you may notice that the roof has a great deal of white noise. This is from the tarps blowing in the high winds of the Columbia River Gorge - which up at the View Point Inn can howl like pandemonium!

(North Elevation Cut - FARO point clouds)

Cloud-to-Cloud with changes in geometry

We also found that the exterior scans were a challenge to register. The high winds blew the vegetation considerably between scans, so the outside geometry was not consistant for cloud-to-cloud registration tools to find matches. To work around this, we chose manual plane and point targets on the building. In a pinch the flag pole out front could be used by extracting a pipe feature in FARO Scene and using it as a target as well. Pipe objects appear to be a heavily weighted registration object and will pull scans together quickly, yet we find that they aren't necessarily best for accuracy, probably because these extracted features are not exactly the same size between scans due to the nature of the tool itself, therefore the extracted centerlines may not be the same. However, once the scans are closer to one another, users can run scan registration again with existing targets or using cloud-to-cloud, usually with some success.

Limitations on FARO X-Series scanner GPS sensors

(West Elevation Cut - FARO)

The GPS on the FARO x330 scanner (used on this project) also has some limitations and doesn't help with the registration process if the scan locations are more than 10m apart. So, while at first glance it might seem that scan technicians could make large leaps between scanning positions and the GPS would save the day with the registration. Not so. Users are still required to keep the scans relatively close to one another to leverage this sensor. Otherwise, an error will appear in the registration process indicating that the GPS is too far away from the GPS reference position. We produced better results by simply unchecking the GPS sensor in the registration process for outdoor scans that were greater than 10m apart. From the manual:

While SCENE uses a plane coordinate system, GPS takes the earth’s curvature into account. Using GPS information for the registration of scans located very far away from one another may affect the registration accuracy as great distances on the surface of the earth are different than on a flat map. To prevent such registration inaccuracies, SCENE defines a GPS reference position for the workspace. GPS information of objects within a certain radius (10.000m) from this reference will be used for scan registration; GPS information of objects further away will be ignored.

We received some GPS signal adjacent to the outside walls of the interior of the building. This helped to align scans adjacent to the exterior scans, which could be seen through the major openings in the now dilapidated Inn.

(Iso-Cut: FARO point cloud looking east)

Also to note, this information is not available in the new FARO Scene interface help PDF. Users will need to switch to the old user interface to access the help PDF that contains this information. That being said, it is also worth noting that there are several tools missing from the new interface which still exist in the old interface. While it may take a bit more experience with FARO Scene to use the older interface, it is more robust for meeting your scan processing needs. It may not be as pretty, but it is very useful.

East point cloud elevation: cut to apex of roof

West point cloud elevation: cut to apex of roof

Floor Plans: Base Point Clouds

Point Cloud - View Point Inn Basement (cut): Wet, Dark and Scary... Floodlights placed under the scanner tripod brought a little color to the situation

Point Cloud - View Point Inn: A very soggy Ground Floor (cut) requiring us to walk on the basement ceiling beams so as not to fall through

Point Cloud - View Point Inn: Top Floors (cut) with rain and high winds buffeting through the open rafters above

Exporting

Exporting the point clouds was accomplished in two ways: first ToPa 3D exported the individual scans and converted these to ReCap. This facilitated the use of 'Mirror Balls' (bubble views) of each scan for PMA to evaluate. The second export was leveraging Project Point Cloud and slicing up the scans using the MultipleClippingBoxes tool in FARO Scene (for more information on this workflow, see our previous blog on Visualizing Point Clouds). The purpose in exporting from clipping boxes was to provide support in the Revit modeling process. Each clipping box contains scans from multiple vantage points and because they are processed through the Project Point Cloud tool, they are color balanced as well. Clipping scans in this way provides some level of resource management for these robust scan files, enabling modelers in Revit to turn off and on point clouds as needed. ToPa 3D also provided a naming convention for the modelers, showing where these clipping boxes were located in relation to one another. 

Exterior clipping boxes for .PTS exporting from FARO Scene

Interior clipping boxes for .PTS exporting from FARO Scene - note that because the interior scans have a higher density of points, we made the clipping boxes smaller to keep the relative file sizes small.

sUAS Documentation

Finally, the entirety of the property was mapped with a DJI Inspire Pro drone. Certainly the most smile-enhancing aspect of this project. Using a combination of the exterior LiDAR (UTM 10N - extracted from our FARO Focus3D x330 GPS sensor data. Note that this coordinate information is important post-processing data for ContextCapture to blend LiDAR with geotagged imagery) and aerial mosaic imagery, we processed the DTM in Bentley's ContextCapture. If you're curious on what sort of computer hardware you'll need to process through this software relatively well, the CPU crunches the aero triangulation (AT) solution and the GPU is mostly responsible for the mesh reconstruction. Bentley Systems has a rating system in the ContextCapture specs that indicates - on some level - the performance you can expect with various graphics cards. For example a GeForce GTX 980 has a 100-110 rating, and the new 1080TI may rate at 200, but for performance you shouldn't necessarily expect double the speed, rather about 1.5x improvement could be expected. 

FARO Point Cloud before sUAS (drone) images applied in ContextCapture

FARO Point Cloud before sUAS (drone) images applied in ContextCapture

FARO Point Cloud and sUAS images combined in ContextCapture. Note that all the occlusions are filled, yet the model remains accurate from the 3D laser scans.

Essentially, the sUAS imagery is "shrink-wrapped" over the FARO point cloud creating a highly detailed and accurate, measurable mesh.

One of the powerful aspects of ContextCapture is that it allows users to align the photos to the LiDAR and fills in the holes where a scan may have had occlusions, creating a more complete mesh, or a more complete point cloud - depending on what the data-type goals of the project are. Also worth mentioning is that ContextCapture can output not just a mesh and point model, but also a KML for Google Earth. Bentley offers a great deal of user-friendly versatility for the design process with this platform.

Here is a brief animation of the initial photogrammetry-based point cloud from the Inspire.

Bringing it all together

The overall goal of this project was simply to create precision as-builts of both the property's primary features; the interior and exterior. However, we at ToPa 3D always seek to add value to the project if time permits. On this project, we accomplished this through the use of 3 sensors - FARO 3D laser scans, Matterport, and Aerial Photogrammetry via drone imagery. All of these sensors produced a point cloud that was measureable from a micro-scale (interior spaces) to the macro scale of the entire property. The Matterport Remote Site Access feature we offer our clients, allows project modelers to gain a real sense of the space for the design process without the limitations of standard architectural photography using traditional methods. Matterport was much faster and less expensive than traditional methods, while offering an exponentially greater number of visual references for the modeling and assessment process.

The Matterport point cloud covered the entire building interior in ReCap format with a file size of only 500mb as opposed to the 16gb interior ReCap scan sizes from the FARO scanner. While PMA primarily relies on the FARO scans for accuracy, to get a quick glance of the entire space in Revit via the lightweight Matterport point cloud is proving very useful. The total Matterport acquisition time was less than 2 hours. Processing was about 4 hours beyond that.

The overall point cloud created from our drone will provide a comprehensive, photorealistic site plan with the supportive accuracy of about 14 FARO scans - scans that, due to project time constraints, did not capture the cliff areas on the edges of the property, road features and other infrastructure items that may be useful down that project road. The total drone acquisition time was less than 30 minutes. Processing time for this number of photos and scans is less than 24 hours (we are awaiting survey control to reprocess and verify accuracy of this relatively new technology).

What's next and a personal note...

The writing on the wall is that reality capture technology will just become faster and more efficient.

Today a colleague in the sound production industry found himself getting worked out of a tech job because the software he spent his adult life mastering is becoming so efficient and easy to use, most people are just doing it themselves now. The same is true with some of the surveying and modeling techologies we use today. However, as long as business owners in this industry are willing to continually innovate, stay creative, and most of all, enjoy the process of learning new things, they have nothing to worry about as we see it. While a gross simplification of what could be called a challenging industry to keep up with, we believe that there will always be new tools in the wheelhouse to play with that firms will not necessarily want to do for themselves. They will still call upon those who are closely tied to what the trends are and what does and doesn't work well.

Years ago, this author studied land surveyor methods (in college as a geology major) and learned the math for COGO digital mapping. For about 10 years, the Land Surveyor Reference Manual was a sort of bible while working on cadastral mapping projects before ArcMap and geodatabases were invented and everything was a clunky coverage file. We used Microstation extensively in the 1990's and early 2000's, writing scripts to convert vector types, colors and weights to coverages in ArcInfo, enduring "Fatal Exception Errors" and sifting through thousands of deeds, surveys and plats to find corners that we could digitize with some degree of accuracy for tax assessor maps. A bit mind-numbing over time. Moving into the world of 3D full time was and remains a very exciting endeavor. Not all advertised hardware produces survey-grade accuracy, but not every project requires this. So there has been a little bit of letting go of my "accuracy OCD" that I was used to for so long. We work with clients to ensure accuracy requirements are met - some are ok with just a Matterport or drone mesh in order to have a general context without a need for tight tolerances, and some need that millimeter accuracy or requires our data be tied with survey control. It's a good mix of math and creativity working at ToPa 3D and soon, we'll be looking to add on more talent to our team thanks to many successful projects.

At the end of the day, I can say there is nothing I would rather be involved with than this industry and the R&D that has become an integral part of it. Finding the right tool for the job has become a personal passion for some crazy reason and it is how I typically spend my Friday, Saturday, and just a bit of Sunday nights (fortunately I have a very loving and understanding partner...)

If you like to dig into details, have a question on something that isn't working like it should or just want to wax poetically about your inner geek, we would love to chat. And of course, if you need an experienced service provider for that next reality capture project, please contact us at: [email protected]

Remember, Orange Sunshine creates a temporary condition, you are still a living, human organism on the planet earth, and you'll be fine in about 5 hours...

ToPa 3D~

Special thanks to Mike Barkasi at Bentley Systems for the support and technical advice on processing data for the View Point Inn.