Not certain what engine they're using, but they're working to open London's museums up to the world as rift/gear VR experiences - so probably unity or UE4.
The other thing that the article doesn't mention which is really neat is that the Shipping Galleries that they picture at the Science Museum are sadly no more - so their work has preserved those galleries in virtual form, and allows them to live on, even when the physical space has been repurposed.
Edit: here's a nice video of the pointcloud of said gallery.
Really cool, I hope they open the data for improved visualizations. Rendering raw colored points leaves to be desired, some kind of geometry extractor would much improve the display.
"The project required 223 laser scans, collected over four days in December. Each scan takes 10 minutes to half an hour to complete. ScanLab also takes photographs, which it uses later to map color onto the point cloud (newer scanners can capture color automatically, but ScanLab prefers its techniques; the vivid color of the group’s scans is part of its uncanny magic). Once the field work is done, software compiles individual scans into a pointcloud. The Mail Rail scan is comprised of more than 11 billion points, consuming over a terabyte of storage. (When the ScanLab guys talk numbers with companies that specialize in data storage and backup solutions, the experts often assume ScanLab is an architecture studio of 150 employees. It has in fact expanded—it’s now a team of six.)"
I wonder if they use AgiSoft's 'PhotoScan' software to bring their photos and laser scans together into one model?
edit: Looks like they probably use the following scanner: http://www.faro.com/en-us/products/3d-surveying Still curious to know what their process is to bring their own photos and the 3d point cloud data to life
I did a bit of work with point-cloud/lidar data during a placement with an engineering company. They were actually driving down highways with LIDAR apparatus attached to a vehicle to build a point clouds of roads/similar. The sheer detail (points per inch/similar) in the point clouds whilst beneficial also requires ridiculous amounts of RAM to do any form of processing. If rendered as points, it's only useful at a distance, as you zoom in it becomes a void with sparse points.
I tried using PCL[1] to convert a point-cloud into a mesh and I seem to recall it using upwards of 6 times the amount of memory of the stored points (before I ran out of memory). I was only there a short time so I never did get round to making it work.
There's a few innovative approaches to this, some of the done in real time. The "histopyramid" approach is used in the demoscene, and it's explained in the Fairlight rendering secrets talk: https://www.youtube.com/watch?v=ZiPGd3DDfMU
It's maybe not suited to billions of points, but it's a decent start.
Is it the difference between raw data and processed? The scan could have things like [I'm guessing] corrected baseline direction of the laserbeam (3-axes: theta, phi, psi??), time of pulse initiation in ns(?), time of pulse reception, colourific data, CRC. Then you process that to create the point cloud?
I made mine using a line laser and a webcam. Its slow, but I can scan decently well, and in color. Of course, the more lasers and cameras, the faster I can scan.
Something like this will have huge ramifications on the ecommerce, travel industry and esp online video, and with a tech like this, they are going to absoultey crush their non-digital counterparts.
Imagine buying a dress using a VR which lets you explore the dress with pin point accuracy and detail. This is the kind of stuff that would help save a tonne of money in returned products for fashion bases e-tailers. It won't be surprising to see these companies being acquired by Amazon, Google, or Facebook to be integrated with their ecommerce, gaming businesses.
14 comments
[ 3.0 ms ] story [ 47.7 ms ] threadThe other thing that the article doesn't mention which is really neat is that the Shipping Galleries that they picture at the Science Museum are sadly no more - so their work has preserved those galleries in virtual form, and allows them to live on, even when the physical space has been repurposed.
Edit: here's a nice video of the pointcloud of said gallery.
https://www.youtube.com/watch?v=gDTbFhFZl9I
I wonder if they use AgiSoft's 'PhotoScan' software to bring their photos and laser scans together into one model?
edit: Looks like they probably use the following scanner: http://www.faro.com/en-us/products/3d-surveying Still curious to know what their process is to bring their own photos and the 3d point cloud data to life
I tried using PCL[1] to convert a point-cloud into a mesh and I seem to recall it using upwards of 6 times the amount of memory of the stored points (before I ran out of memory). I was only there a short time so I never did get round to making it work.
[1]http://pointclouds.org/
It's maybe not suited to billions of points, but it's a decent start.
That's about 90 bytes per point. Seems a lot.
That's 15 bytes. So.... I'm 600% more optimized than Faro? I think someone's off on their numbers.
Imagine buying a dress using a VR which lets you explore the dress with pin point accuracy and detail. This is the kind of stuff that would help save a tonne of money in returned products for fashion bases e-tailers. It won't be surprising to see these companies being acquired by Amazon, Google, or Facebook to be integrated with their ecommerce, gaming businesses.