What I learned from scanning Alcatraz – Part 2

Article by Dr Jeremy Sofonia

You’ve done the planning, now in the second of this two-part series, Dr. Jeremy Sofonia, Emesent’s Senior Technical Specialist, talks about some of the tips and tricks that helped ensure successful data capture when mapping notorious US island prison, Alcatraz.

Part 2 – The Data Capture

Check as you go

The opportunity to scan Alcatraz was a rare opportunity compressed into a short timeframe so, if we missed something or a data set didn’t turn out the way we hoped, it wouldn’t be easy to go back. This made it all the more important to take a quality assurance approach.  We easily captured over 100 Hovermap scans throughout the whole project and found it much more efficient to check that what we’d captured was true and correct before we carried on. We used the visualization tool in Emesent Aura to load the point cloud and quickly identify if a scan had any issues. This allowed us to make a decision as to whether or not it was faster to try to reprocess it with different settings, or to simply scan it again. 

By processing as we went it also allowed us to keep up with the sheer volume of data without creating a huge backlog to address. 

Close the loop

Emesent’s SLAM algorithm is extremely robust, but a slip in the point cloud can happen. For example, if I was in a small space, maybe I blocked Hovermap’s field of view, and it wasn’t able to link up this room to the previous room. Visualizing the point cloud by time is a very quick and easy way to see if the point cloud has any issues. Coloring the point cloud by time is an easy way to confirm that there are no slips, particularly if the scan had a ‘closed loop’ (starting/ stopping the scan in the same area).

Colored by time objects will be in different colors on top of each other and there shouldn’t be two fire extinguishers where there’s one, for example. This allowed us to see if there was a problem very quickly, validate the scan and move on. 

Loop closure is a good thing but anytime you can cross your own path, or rather Hovermap crosses the same area, is even better.  This allows Emesent’s SLAM algorithm to compare its current location estimate, and correct any differences to what was scanned before. In short, loop closures are good whether big or small.

Embrace the versatility – use the best option

We were tasked with making the most complete model possible so it was important to leave no stone unturned. However, there were some ‘no go’ areas for personnel due to the presence of asbestos and lead. To use PPE and respirators would have required a lengthy approval process and would have still exposed personnel to risk. 

Instead, we used Emesent’s Hovermap LiDAR mapping system integrated with the Boston Dynamics Spot ground-based robot. This allowed us to operate Spot remotely, safely outside the ‘no go’ area and monitor the data in real time as it was captured. 

Color capture

In my three weeks on Alcatraz I probably captured more colored datasets than in my entire career combined. So I did a lot of experimenting on the most efficient techniques to use when scanning with color to ensure full data capture.

How do I enter the room? How do I cross the room? Is it better to walk down the middle of the room or do I hug the walls? Do I go to the corners? Do I walk in a circle? I found that the most effective technique was to scan along the walls, from corner to corner to corner, pointing the GoPro toward the center of the room as well as floors, ceilings and adjacent walls – and it didn’t matter if it was a large room or a closet. I would still get in the closet and try to go from corner to corner to corner to make sure I was capturing every possible angle, because you need that information to be able to generate a complete colorized point cloud with few shadows.

On Alcatraz, I was using a standard GoPro.  Since then, Emesent has released 360 point cloud colorization supported by the GoPro Max 360 camera.  Ambient light is equally important but the need to “paint” the room with the camera is greatly reduced as both Hovermap and the GoPro are imaging in near 360°.  The corner-to-corner approach is less important but care is still needed to cover the room and around objects to ensure good coverage and minimize shadows.

Regardless of the GoPro used, it’s well advised to transfer its files into the scan folder as soon as possible.  As they don’t have a unique name, it’s very easy to get them confused when capturing several scans back to back.  Assigning them to their respective scans, sooner rather than later, ensures they are properly allocated and saves time replaying videos later to figure out where they belong.

Slow and steady

When coloring with Hovermap and the Go-Pro, slow and steady is often best. Take a Goldilocks approach using speed  instead of temperature –  not too fast, not too slow.

Too fast can blur imagery, too slow is just adding processing time with no benefit. This is still best advice but some new tools in the latest release of Emesent’s Aura processing software really improve on this by pulling imagery based on Hovermap’s movement rather than time. This generally means that only new imagery is captured without impacting results, while at the same time improving processing times without additional, unnecessary, information.

Visit emesent.com/alcatraz to hear the stories behind the scans.