This morning, July 13, 2009, LizardTech unveiled LiDAR Compressor. This application allows consumers of LiDAR data the same benefits LT has provided raster consumers for over 10 years. The press release describes some of the product’s overall capabilities and benefits. I was privileged to be the project manager for this effort and in this post will describe the product from an engineering perspective.
At the lowest level, the application leverages our experience doing wavelet-compression. Our specific approach is the subject of a patent application and I won’t attempt to describe it in detail. However, a concise summary is that we do a 1D wavelet transform on each of the channels of interest (x, y, z, intensity, etc) and then use the same compression techniques we use in our MrSID and JP2 implementations (bitplaning, entropy encoding, appropriate quality-weighting of the bitplanes).
At the SDK implementation level, we depend on several packages from the OSGeo stack. We use liblas to read data out of the LAS files. LT is proud to be the first (to my knowledge) commercial application of this industrial-strength library. (Regular readers of our blog may recall that last March we supported the OSGeo Code Sprint in Toronto; liblas was certainly part of our interest there!). Liblas, in turn, uses libgeotiff to decode the CRS information in the LAS files and the now venerable GDAL to translate that into a WKT that the rest of us can understand. The result is a C++ SDK available to our application … and soon to yours too (decode only). We’ve got distributions for both Windows and Linux which will be available on our developer’s site any day now.
At the application level, the underlying technology has changed as well. Our flagship product, GeoExpress, is written in managed C++ using the Windows Forms library. Our most recent products (GeoViewer and Lidar Compressor) are built in C# using WPF. Our experience was that this change allowed much greater developer productivity and the creation of more flexible (not to mention prettier) applications. For example, I suspect we would not even have attempted a ListView with dropdown boxes in the header (all dynamically populated based on the selected input text file — see Figure 1 below) if we had to implement it without XAML support:
Figure 1: Importing a text file.
Additionally, the application makes extensive use of the WPF/.NET threading support. This allows the application to achieve two design goals: (a)it provides first class support for text-based LiDAR files and (b) it can run several jobs at once.
Before we can do any kind of processing (e.g. compressing, viewing) we need to tell the file’s reader how many points there are and what the overall extent is. For LAS files (and MG4 files) this is no big deal because everything we need to know is stored explicitly in a compact header. For text files, it’s not stored at all and must be calculated which involves reading the entire input file. This “initialization” (see figures 2 and 3 below) can take a long time and needs to happen in the background.
Figure 2: Text file initializing (in the background, so the application is still responsive.)
A few minutes later …
Figure 3: We’ve read the 321 Mb text file and now know the extent and number of points.
Running several jobs concurrently in and of itself is not a hard problem. The difficulty comes when we consider how to provide feedback (progress bar, log messages), the ability to cancel and deal with failures. For historical reasons, GeoExpress runs its compression work in a separate process. This provides excellent isolation (uh,… that’s part of the ‘history’) but complicates tight integration with the parent application. This is part of the reason we could not include the ability to run concurrent jobs in GeoExpress 7. Lidar Compressor uses WPF’s built in BackgroundWorker class to manage the worker threads. This vastly simplifies the integration tasks. The result is an application that leverages the multi-processor capabilities of modern hardware as well as a UI that is simple to understand and use (see Figure 4 below).
Figure 4: Five jobs running simultaneously.
Everything the user sees is pure WPF except for one thing: the viewer. The viewer is a WPF wrapper around DirectX. There’s a lot of math that goes into manipulating 3D objects. DirectX provides an API to do this for us and, very significantly, runs it on the graphics card (i.e. no burden on the CPU). So, the good news is that rendering is blazing fast. The bad news is that DirectX 9 D3D (which is what we use) gets very finicky when it comes to older OSs (that is, XP), drivers and hardware. We were not able to get a really satisfying resolution to when DX would fail and why, but we were able to guard against it so that the application handles the failure gracefully.
Checkout the free trial and let us know what you think.
Posts
