Another week, another conference. I have to say, I have been rather excited about this workshop as it brings together all the top dawgs of 3D around Europe at a location which I feel comfortable in and isn’t too far away 🙂 Having arrived the night before at a lovely B&B in Romsey, I suited up, armed every pocket with business cards and headed to OS.
In the morning there were a couple of warm-up sessions, if you will. First it was a demo of UVM/CityGRID but I will discuss that later in Gerald’s talk. The second session was a demo from Virtalis on their GeoVisionary Software. Virtalis is a VR company with 38 staff, predominantly software developers. They have a “strategic” relationship with BGS for many years. Now GeoVisionary is an impressive piece of kit in that it is capable of render very large 3D datasets incredibly quickly. It also allows for collaboration across networks with only 1-2 seconds latency. With GeoVisionary 3 (due March 2015), there are also options to stream directly from databases such as Oracle. Incredible! The set up for one of these systems was ridiculous though and they avoided naming the price of the software. One thing I didn’t like was the fact that they marketed it as “oh it’ll run on a laptop even with 2GB of RAM”….. but they didn’t even mention the massive graphics card and processor that is also installed in the laptop. But otherwise, solid 3D viewer. Shame the preprocessing takes weeks at the national level…
====================SESSION ONE: The 3D Capture Process====================
Automatic Texturing of 3D Citymodels from Aerial Images – Gerald Forkert, UVM Systems, Austria
I love building textures. And for me, it is one of the most important part of a 3D city model. Gerald explains how it is not the geometry, but the textures that help people navigate/identify buildings. Using 80% along, 60% across overlap imagery, and both nadir and oblique images, Gerald was able to create pretty convincing automatically textures 3D city models. While the nadir images provided great roofs, the obliques gave us higher facade resolution. Occlusion analysis avoided mistexturising and the high overlap gives us lots of options. At around 50EUR per sqkm plus 1000EUR start up costs, it seems like a reasonable option for texturising your 3D buildings!
Semantic and geometric enrichment of 3D geo-spatial building models – Jonathan Slade, School of Computer Science & Informatics, Cardiff University
Having worked before/am currently working closely with Jon, I was keen to see what he has been up to since we last met. His experiments on template matching using normalised cross correlation and non-maximum suppression is very interesting. Not fool proof as of yet, but clearly lots of opportunities there to be exploited. Very exciting stuff.
Multi-ray photogrammetry: A rich dataset for extraction of roof geometry for 3D reconstruction – Andrew McClune, Newcastle University, UK
Another fellow OS PhD colleague, Andrew’s work focuses on roof geometry extraction. Some very clever photogrammetry techniques here to ascertain roof shape combining scan line segmentation and canny edge detection. Also excited to see how this turns out!
Our experiences in generation of countrywide 3D building models: A focus on process chaining – Markus Braun, MOSS, Germany
I hadn’t heard of MOSS before, but it is a Germany company with 60 employees across 3 offices. Using a mixture of automatic and manual techniques, they have around 50 mil buildings in Germany, of which around 30 mil are automatic. 65% average automatic detection rate.
Designing 3D GI for Navigation Using Google Glass – Kelvin Wong, Department of Computer Science, UCL
Hadn’t heard of this guy before, but what a fantastic speaker. 10 out of 10.
The building blocks of user-focused 3D building data – Isabel Sargent, Ordnance Survey
Izzy really brought it home for me with regards to the difficulties that NMAs face when it comes to 3D. The fact that NMAs tend to be larger, cumbersome beasts who are capable of producing world class research in 3D – but are also blighted by its own corporate inertia to implement it quickly. Research into 3D has existed in OS since 1990s – references to dolly the sheep, furbies and old school Nokia struck a nostalgic note.
Over the years, OS has been gathering information on customer needs to help identify the 8 characteristics of 3D building geometry. But what really made a point for me was the fact that there were 7 types of building height attribute, not all available, but still… 7 types! It shows how difficult to represent 3D purely as 2D height attribute.
Izzy also showed some of the latest research on using machine learning for roof shape classification – incredibly cool and very pretty too.
It seems that 2.5D serves most of its customers’ needs, but has the customers’ needs changed over the last decade? Well I hope to find that out soon!
The process to generate a nationwide 3D virtual model of the Netherlands – Sander Oude Elberink
Sander was here to tell use about 3DTOPNL, a project led by Kadaster with UTwent, TU Delft, Geodan and Conterra. It is a 3D version of the 2D national topo map, combining Lidar point clouds and national height dataset. Lots of pre and post processing in FME, 3D calculation in open source tools and some clever techniques to tackle issues at polygon boundaries and also water. With 15 mil polygons, a super computer was required to do the processing. 3DTOPNL is now open (sinec Nov 2014) and the use-cases are being logged. I’ll be interested to see who picks up this dataset!
Using spatial databases to efficiently manage 3D objects at a national level – Hans Viehmann, Oracle
Some great examples from Hans on the application of spatially enabled databases for 3D data at the national level. Hans sums up the way to scale from 2D to 3D at a national level with three succinct options:
“1.Use more sophisticated software
2.Buy faster computers
Without being too flippant, that does seem to be the case. With numbers frequently ending in “-illions” when referring to national datasets, I’m glad that Oracle are testing different storage methods to maximise query efficiency at a national scale.
GIS for 3D Object Generation, Modelling, and Exploitation – Paul Hardy, ESRI Inc.
ESRI seems to be embracing 3D – I am just waiting for 10.3 to come out with ArcGIS Pro for me to decide whether they are really making good progress or not.
All in all, an awesome two days. Lots learnt, friends made, sandwiches eaten. I look forward to working with many of these EuroSDR folks closely in the coming years…