TIN Terrain
TIN Terrain is a command-line tool for converting heightmaps presented in GeoTIFF format into tiled optimized meshes (Triangulated Irregular Network) with different levels of details.
Check out heremaps/quantized-mesh-viewer for examples of rendering output in Cesium.js and Three.js.
Note: This is experimental code, expect changes.
Features
- Takes GeoTIFF with heightmap as an input
- Transforms heightmap into a TIN mesh with a given max-error parameter and outputs into
.objformat - Transforms heightmap into tiled TIN mesh for a given zoom range and outputs tiled pyramid into quantized-mesh-1.0 terrain format
Let us know what you think!
You can help us a lot to prioritize issues in tin-terrain if you submit a short anonymous survey
Installation
You can build and run this tool either with Docker or directly on your system. Building with Docker is easier and works on macOS, Linux, and Windows.
Detailed instructions for each platform are provided below.
Building with Docker
The provided Dockerfile builds the TIN Terrain executable on Ubuntu Linux.
To build the container, execute:
./build-docker.sh
To run TIN Terrain from Docker, type e.g.
docker run -v [local data directory]:/data:cached --name tin-terrain --rm -i -t tin-terrain tin-terrain ...
where [local data directory] is the folder that contains your DEM files and will receive the output files. This will be mapped to /data in the Docker instance, so you should use:
docker run -v [local data directory]:/data:cached --name tin-terrain --rm -i -t tin-terrain tin-terrain dem2tin --input /data/... --output /data/...
Alternatively, use
docker run -v [local data directory]:/data:cached --name tin-terrain --rm -i -t tin-terrain bash
to open an interactive shell which lets you execute tin-terrain and access /data. Any files not stored in /data will be lost when closing the session.
Prerequisites & Dependencies
If you choose to compile and run "TIN Terrain" directly on your system, please note that the following packages must be installed on your system:
- The C++ standard library from the compiler (tested with
clang9.1.0andgcc7.3.0) - Boost (BSL-1.0)
program_options,filesystem,system(tested with version1.67) - GDAL (MIT)
gdal,gdal_priv,cpl_conv(tested with version2.3)
TIN Terrain also downloads some source code of 3rd party libraries during the CMake configure phase:
You can disable this behaviour by passing the -DTNTN_DOWNLOAD_DEPS=OFF option to CMake when generating the project/makefiles. If you do so, you have to download dependencies yourself and also pass the TNTN_LIBGLM_SOURCE_DIR and TNTN_LIBFMT_SOURCE_DIR variables to CMake as well as TNTN_CATCH2_SOURCE_DIR if you want to run the tests.
See download-deps.cmake for detailed version info.
Building on macOS
- Install dependencies, preferably with homebrew:
brew install boost brew install cmake brew install gdal - Create an XCode project:
mkdir build-cmake-xcode cd build-cmake-xcode cmake -GXcode path/to/sourcecode/ open tin-terrain.xcodeproj - Build the TIN Terrain target.
The resulting binaries will be in the Debug/Release subdirectory.
To run the tests, build and run the tntn-tests target.
Possible linking issues with GDAL
If another version of GDAL is present on your machine, the FindGDAL.cmake provided with cmake might not be able to properly detect the newest version of GDAL you installed through homebrew (or any other prefered method).
Whether this is the case, can be easily detected, if the linking step in the compilation errors on linking GDAL.
Therefore you might need to guide cmake to the right location of the GDAL:
cmake -GXcode -DGDAL_LIBRARY=PATH/TO/GDAL/LIB path/to/sourcecode/
e.g
cmake -GXcode -DGDAL_LIBRARY=/usr/local/Cellar/gdal/2.3.1_2/lib/libgdal.dylib path/to/sourcecode/
Building on Linux
- Install dependencies, e.g. on Ubuntu:
apt-get install build-essential cmake libboost-all-dev libgdal-dev - Create Makefile:
mkdir build-cmake-release cd build-cmake-release cmake -DCMAKE_BUILD_TYPE=Release path/to/sourcecode/ - Build the
tin-terraintargetVERBOSE=1 make tin-terrain
The resulting binary should then be ready. To run the tests, build and run the tntn-tests target:
- Recreate Makefile (and set TNTN_TEST=ON):
cd build-cmake-release cmake -DTNTN_TEST=ON -DCMAKE_BUILD_TYPE=Debug path/to/sourcecode/ - Build the
tntn-teststargetVERBOSE=1 make tntn-tests
Usage
The tin-terrain command-line tool has a few subcommands. You can run tin-terrain --help to see all available subcommands.
$ tin-terrain --help
usage:
tin-terrain [OPTION]... <subcommand> ...
Global Options:
-h [ --help ] print this help message
--log arg (=stdout) diagnostics output/log target, can be stdout,
stderr, or none
-v [ --verbose ] [=arg(=1)] be more verbose
--subcommand arg command to execute
--subargs arg arguments for command
available subcommands:
dem2tin - convert a DEM into a mesh/tin
dem2tintiles - convert a DEM into mesh/tin tiles
benchmark - run all available meshing methods on a given set of input files and produce statistics (performance, error rate)
version - print version information
Projections
The tin-terrain tool requires your datasets to be in the Web Mercator projection (EPSG:3857). If your datasets are not in this projection, you can quite easily reproject your datasets with gdalwarp, e.g.:
gdalwarp -t_srs EPSG:3857 ned19_n37x75_w122x50_ca_goldengate_2010.img ned19_n37x75_w122x50_ca_goldengate_2010_mercator.tif
Creating a single mesh/TIN
You can use the dem2tin subcommand to convert a raster heightmap into a single big mesh/TIN in the OBJ format.
You can see all available options by running tin-terrain dem2tin --help.
$ tin-terrain dem2tin --help
usage:
tin-terrain dem2tin [OPTIONS]...
dem2tin options:
--input arg input filename
--input-format arg (=auto) input file format, can be any of: auto, asc, xyz,
tiff
--output arg output filename
--output-format arg (=auto) output file format, can be any of: auto, obj,
off, terrain (quantized mesh), json/geojson
--method arg (=terra) meshing method, valid values are: terra, zemlya and dense
--max-error arg (terra & zemlya) maximum geometric error
--step arg (=1) (dense) grid spacing in pixels
methods:
terra - implements a delaunay based triangulation with point selection using a fast greedy insert mechnism
reference: Garland, Michael, and Paul S. Heckbert. "Fast polygonal approximation of terrains and height fields." (1995).
paper: https://mgarland.org/papers/scape.pdf
and http://mgarland.org/archive/cmu/scape/terra.html
zemlya - hierarchical greedy insertion
reference: Zheng, Xianwei, et al. "A VIRTUAL GLOBE-BASED MULTI-RESOLUTION TIN SURFACE MODELING AND VISUALIZETION METHOD." International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences 41 (2016).
paper: https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLI-B2/459/2016/isprs-archives-XLI-B2-459-2016.pdf
dense - generates a simple mesh grid from the raster input by placing one vertex per pixel
For example, you can run the following command to convert "ned19_n37x75_w122x50_ca_goldengate_2010_mercator.tif" to a single big mesh, using a max-error parameter of 2.0 meters:
tin-terrain dem2tin --input /data/ned19_n37x75_w122x50_ca_goldengate_2010_mercator.tif --output /data/terrain.obj --max-error 2.0
After running this command, you should see a "terrain.obj" output file in the same directory. This OBJ file contains the generated mesh.
The max-error parameter specifies the vertical error allowance in meters, so a smaller max-error parameter results in more triangles in the output mesh, better mesh quality, and longer running time.
Creating a pyramid of mesh/TIN tiles
The dem2tintiles subcommand takes a raster heightmap as the input, and creates tiled TIN mesh for a given zoom range. The output tiles can be either in the OBJ format or in the quantized-mesh-1.0 terrain format.
You can see all available options by running tin-terrain dem2tintiles --help.
$ tin-terrain dem2tintiles --help
usage:
tin-terrain dem2tintiles [OPTION]...
dem2tintiles options:
--input arg input filename
--output-dir arg (=./output)
--max-zoom arg (=-1) maximum zoom level to generate tiles for. will
guesstimate from resolution if not provided.
--min-zoom arg (=-1) minimum zoom level to generate tiles for will
guesstimate from resolution if not provided.
--max-error arg (terra or zemlya) max error when using
--step arg (=1) (dense) grid spacing in pixels
--output-format arg (=terrain) output tiles in terrain (quantized mesh) or
obj
--method arg (=terra) meshing algorithm. one of: terra, zemlya or dense
For example, you can run the following command to convert "ned19_n37x75_w122x50_ca_goldengate_2010_mercator.tif" to a pyramid of mesh tiles.
tin-terrain dem2tintiles --input /data/ned19_n37x75_w122x50_ca_goldengate_2010_mercator.tif --output-dir /data/output --min-zoom 5 --max-zoom 14 --output-format=terrain --max-error 2.0
When this command finishes running, you should see an output folder containing all the mesh tiles, organized into a pyramid of subfolders.
├── 10
│ ├── 163
│ │ ├── 627.terrain
│ │ └── 628.terrain
│ └── 164
│ ├── 627.terrain
│ └── 628.terrain
...
The folder structure follows the map tile convention: Z/X/Y.terrain.
These mesh tiles can then be easily served from a webserver and be consumed by frontend applications for purposes such as terrain visualization.
Sample Datasets
When you enable the TNTN_TEST and TNTN_DOWNLOAD_DEPS options in the CMake configuration, a few sample datasets will be downloaded into the ${CMAKE_SOURCE_DIR}/3rdparty/ folder.
For example, you will find the crater lake dataset in the ${CMAKE_SOURCE_DIR}/3rdparty/craterlake folder. This dataset is created and maintained by the U.S. Geological Survey (USGS) and can be downloaded from http://oe.oregonexplorer.info/craterlake/.
To generate a mesh from this dataset, you need to reproject it to the Web Mercator projection first, using the gdalwarp command-line tool which comes with your GDAL installation:
gdalwarp -t_srs EPSG:3857 -r cubic -of GTiff -ot Float32 dems_10m.dem dems_10m.tif
Then you can run tin-terrain on this reprojected GeoTIFF file to create a mesh:
./tin-terrain dem2tin --input dems_10m.tif --output terrain.obj --max-error 2.0 --method terra
License
Copyright (C) 2018 HERE Europe B.V.
See the LICENSE file in the root of this project for license details.
Papers
Some algorithms are based on ideas from:
- Garland, Michael, and Paul S. Heckbert. "Fast polygonal approximation of terrains and height fields." (1995).
- Zheng, Xianwei, et al. "A VIRTUAL GLOBE-BASED MULTI-RESOLUTION TIN SURFACE MODELING AND VISUALIZETION METHOD." International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences 41 (2016).
