[talk-ph] Yahoo!'s satellite imagery in Makati CBD is not good enough

Rally de Leon ralleon at gmail.com
Thu Apr 9 18:22:45 BST 2009


Eugene,

Try to can gain access to adjacent  villages (eg. bel-air, urdaneta village,
san antonio village, san lorenzo) where there are no tall buildings (eg.
less than 500m-1km away from CBD). The technique is to get multiple gps
tracklogs (carefully taken, slowly driving in the middle of the roads) back
and forth, criss-crossing these narrower roads of the residential
subdivisions. You'll get fairly accurate (unobstructed) gps traces here, and
they're so near your target area (the CBD). The resulting tracklogs in and
around these villages will be your reference lines for correcting the traces
on the CBD itself.

First step: trace over the satellite images inside these residential
subdivisions together with CBD (up to the portion of the anomalous image
area only), then upload to OSM.

On a separate layer in JOSM, drag and calibrate these traces (both adjacent
residential subd and the nearer portion of CBD) over the layer of your gps
tracklog of the subdivision. Save your work.

Use this same technique on the opposite portions of the CBD (using another
adjacent subdivision as reference), until all portions of your
gps-calibrated traces meet along the anomalous satellite images of the CBD.
Use a different color of line (per layer) of your CBD portions, so they
won't visually mix up when you try assemble and combine all into a single
layer later.

Once you combine the CBD traces into a single layer in JOSM, get the best
compromise (by redrawing and stitching your different CBD portions) using
the natural-geometric shapes and alignments of the roads. If it looks good
(not crooked), then it must be fairly accurate.

What you'll get is a very good gps-aligned CBD map-trace, without
necessarily trying to collect gps tracklogs on the urban canyons of the CBD
itself.

The only problem is how to get inside the exclusive subdivisions on a
mountain bike or a car (with your gps), without alarming the guards. Roaming
around is normally prohibited, unless you are a resident of the place. :-)

NOTE: on your suggestion to use "top of the building" for gps-reference
waypoints, it won't work. You'll notice that the satellite images of most
tall buildings are not vertically straight (buildings look like 3-D images
swaying to one side). It's roof-top does not necessarily correspond to the
building's footprint at street-level (on the satellite image). Therefore,
GPS data must be taken only at street level (if you will be using these data
along side your satellite traces). On similar case, you must also be careful
in tracing roads near a steep hill or mountain. Tracing roads with
"greatly-varying altitudes" won't give you accurate data.  I noticed this
problem years ago when my "carefully-taken gps tracklogs" from the lowlands
of Tanay Rizal (along Sampaloc Road) zig-zaging all the way up  to the upper
portion of Sampaloc Tanay, won't fit the road shape on Google Earth's
satellite images. Same case for OSM tracing. I realized that I must chop my
road traces into shorter portions, then align them independently to the
actual gps tracklogs every few hundred meters or kilometer (depending on how
rolling the terrains are). Multiple GPS tracks are still the best
reference/s for actual positions (and offsettings); but as to actual "road
shapes and geometries" of an area with the same/common road elevation, our
"handheld gps" tracklogs can't beat traces over satellite images.
What-you-see-is-what-you-get :-)

BTW, this technique is also good for tracing (and calibrating) very wide
road (10-lanes) such as EDSA or C-5 or NLEX. You can't legally slow down and
choose a lane in the multi-lane EDSA to collect reference tracks in tracing
EDSA itself or it's vicinity. It is best to get the gps reference tracklogs
from narrow (parallel and perpendicular) roads on the sides of EDSA (or
C-5). Again, never use tracklog taken from very wide roads like EDSA,
Commonwealth, NLEX and alikes, as reference tracklogs for your city map
(satellite) traces, unless we have no choice like in the rural areas (eg.
SCTEX traces - with no side roads nearby)

Cheers,
Rally

On Thu, Apr 9, 2009 at 8:40 PM, Eugene Alvin Villar <seav80 at gmail.com>wrote:

>
>
> On Thu, Apr 9, 2009 at 7:31 PM, D Tucny <d at tucny.com> wrote:
>
>> 2009/4/9 Mike Collinson <mike at ayeltd.biz>
>>
>> At 11:26 AM 9/04/2009, Eugene Alvin Villar wrote:
>>> >Hi all,
>>> >
>>> >I was doing some cleaning up and some building mapping in the Makati CBD
>>> area and I noticed that the satellite imagery in Yahoo! (provided by GeoEye)
>>> has some really bad stitching (multiple satellite imagery were stitched into
>>> one "seamless" mosaic). One particularly bad example is that Rufino St.
>>> (Herrera) is broken along Ayala Avenue. The stitching of the imagery seems
>>> to be along Ayala Avenue and Buendia. If you'll check out the parts of
>>> Buendia near the RCBC Plaza, you'll notice that there are two shadow images
>>> of Buendia there.
>>> >
>>> >I think this means that the data in the Makati CBD area might be quite
>>> off in some parts. I think that we need to supplement this with some really
>>> good GPS traces (the existing uploaded GPS traces are quite noisy) but the
>>> problem is the urban canyon effect that makes GPS a bit ineffective in this
>>> area.
>>> >
>>> >What do you guys think?
>>>
>>> I tried GPS mapping in Makati CBD before the imagery was available and
>>> found the GPS quality just awful, you are very polite :-) .  I had a similar
>>> problem in Sydney CBD coupled with very oblique aerial imagery that obscures
>>> many of the roads.  The best solution I came up with was to wander about and
>>> take lots of digital photographs down streets and then iteratively edit the
>>> map to get relative position looking right against the photos with the few
>>> spots of good imagery and GPS as a control for absolute position.
>>>
>>> Mike
>>>
>>
>> If it's possible to get onto the roof of some of the buildings, you could
>> get some longish term averaged GPS fixes, and, if you can get ontop of some
>> of the tall buildings, some aerial photos :)
>>
>> The urban canyon effect is going to pretty much rule out getting 'really
>> good GPS traces' as you'll notice that doing everything you can to get good
>> signal, the trace will still be all over the place... Maybe in a few years
>> when GPSrs are better and you can use Gallileo and the US GPS together then
>> you'll be able to get something half decent, but, for now, creative thinking
>> and as Mike says, plenty of photos are probably the only way to get a
>> reasonable level of accuracy...
>>
>> d
>>
>
> I actually thought about going to the rooftops of selected buildings to get
> absolute fixes for these points. Then lay out everything else relatively.
> But knowing about paranoid security, it'll be hard to access the rooftops of
> the right buildings. :-P
>
> I'll do some research on which buildings are generally access=permissive to
> the rooftop. :-)
>
> Eugene / seav
>
>
> _______________________________________________
> talk-ph mailing list
> talk-ph at openstreetmap.org
> http://lists.openstreetmap.org/listinfo/talk-ph
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/talk-ph/attachments/20090410/2d7ea11c/attachment.html>


More information about the talk-ph mailing list