[gdal-dev] GSoC GDAL2Tiles: Opacity for Google Maps Overlay, Embedded Google Earth, Tile generation

[Klokan Petr Přidal]

Hi folks,

1. What did I do recently:

- Update of the Google Maps based viewer. I designed my own Opacity
Control for Google Maps - so overlay has a changeable opacity on the
web. The HYBRID transparent labels are over the custom overlay (a lot
of work :-().
- Update of the Google Earth Plugin viewer (embedded in webbrowser),
so it is loading the KML SuperOverlay for tiles.

look at: http://www.klokan.cz/projects/gdal2tiles/overlay/googlemaps.html

- OpenLayers based viewer implementing the same functionality (the
opacity control is not finished yet) but the overlay is displayed on
top of all possible maps (and can be combined with custom WMS source).
OpenLayers supports now all Google Maps, Microsoft Virtual Earth Maps,
Yahoo Maps, OpenStreetMap and OpenAerialMap.

look at: http://www.klokan.cz/projects/gdal2tiles/overlay/openlayers.html

- A set of python scripts for testing of the quality of the tile
generation for new version of GDAL2Tiles. To choose best quality/speed
combination without dependency on external libraries. In the end I am
using gdal.AutoCreateWarpedVRT() for reprojecting the supplied raster
to Spherical Mercator, then requesting a window by ReadRaster(). This
window is stored in memory (MEM driver) and has bigger size then the
needed tile. The tile itself is created by scaling down this bigger
memory dataset to the final size by gdal.RegenerateOverview()
('average' algorithm) and then by calling CreateCopy to final PNG
file.

*** I hope this is the best way how to generate high-quality
reprojected tiles from a supplied raster. If you have some ideas
regarding this, please let me know...

This approach is used only for maximal resulution tiles. The higher
level of the tile pyramid (overviews) are going to be done the same
way like gdal_retile.py do, e.g. by reading existing 4 tiles and
scaling down to the one with the same geographical extent. I plan to
use again gdal.RegenerateOverview(), but I there is a way how to do
that with gdal.ReprojectImage() - even for the non-georeferenced
rasters. When calling ReprojectImage you can choose scaling algo
('bilinear','cubic','lanczos',..), but I prefer to use Overview's
'average' algorithm.
This seems to be fastest way and it guarantee the best results even on
the top of pyramid which are not achievable by other technique.
It is fastest also because the source raster can be Warped VRT.

- I was playing with the new GDAL 1.6 (trunk) functionality for  which
is replacing pct2rgb.py (gdal_translate -expand rgb|rgba). When used
in combination with VRT driver it perfectly serves the raster files
which has palette inside (typically NOAA BSB, USGS, some other TIFFs).

The code is published in my GDAL SVN sandbox.

2. What do I plan to do next week:

- Merge the developed functionality with gdal2tiles.py from gdal-trunk svn.
I plan to add parameters:
-s_srs 'source projection'   - for specifying a spatial reference
system for file which does not have it embeded
-zoom number|from-|-to|from-to -for selecting zoom levels (of Google
Maps/SphericalMercator pyramid) which will be generated from raster
-format 'mercator'|'geodetic'|'simple' - for selecting the tms profile of tiles
-forcekml will have extended functionality, as it will generate kml
for tiles in any projection (including Spherical Mercator) - it is not
cartographically correct, as you are using for example mercator raster
on top of Platte Carre, but it is a requested feature as then you need
only one set of tiles for both web viewer overlay (Google Maps like)
and 3D overlay (Google Earth). The cartographic shift is disapearing
as you are zooming closer. I think I will include some kind of
warning.

- Finish the templates for gdal2tiles and use them in code of the utility.

I guess the documentation and submitting of the new opacity controls
to OpenLayers trac / Google Maps API Examples will wait after
pencil-down deadline.
I plan to create a nice website about Spherical Mercator projection -
to explain how the tiling in Google Maps and similar services runs. It
will use the diagrams and graphic from my poster:
http://www.klokan.cz/projects/gdal2tiles/gdal2tiles-poster.pdf

3. I blocked on:

- OpenLayers synchronous AJAX loading of XML. Right now the generated
web viewers has the bounding box of raster encoded inside the
JavaScipt.
I am afraid I will not implement the loading of Metadata
(tilemapresource.xml) dynamically as it is hard to do (overlay will
appear later) and there are issues with XSS (you can't load XML from
another server then the html viewer is without setting up of a cgi
proxy)

Klokan