[OSM-talk] Introducing SwiftAddress, an highly efficient way of collecting housenumbers

Matija Nalis mnalis-openstreetmaplist at voyager.hr
Wed Jan 27 01:46:47 UTC 2021


On Mon, 25 Jan 2021 17:48:55 +0100, Simon Poole <simon at poole.ch> wrote:
> Am 25.01.2021 um 05:55 schrieb Matija Nalis:
>
>> - Vespuccy
> It's Vespucci, as in Amerigo Vespucci the person the continent is named  after.

Oops, sorry. 
I did manage to spell it correctly in the prologue but messed up afterwards :(

>>     PROS:
>>       - AI for address guessing (pays off if loooong streets, but not in other cases)
> It's not AI, just extrapolating from the geometry. Number prediction 

Well, AI is not much more than correlating some information and making the
educated guess... So any sufficiently advanced algo is same as magic/AI 
to me :)

> will obviously not work particularly well for "randomly" assigned 
> numbers, however avoiding suggesting the same number more than once and 
> automatic addr:street detection will provide some benefit in any case.

Yes, there are some random streets like that, but my bigger problem are
urban streets with a lots of small forks with just a few houses each. 
Also, they often have their own name (Which is similar, but not same,
as the name of main road).

>>       - some consistency checking
>>       - can draw building polygon (but at a loss of ease of use)
>>       - can kind-of work offline if you know and prepare mapping area in advance...
> Just ask for me to add your area of interest to 
> http://mapsplit.poole.ch/ and then you can use daily updated offline data.

I'm mapping Croatia (OsmAnd full map is about 150MB), so it would be great
if you could add it!

> If you want to use background imagery offline you can use MBTiles.

Oh, I wasn't even counting on aerial images being available offline; bitmap
tile data is too prohibitive for larger areas - there is no way I could put
whole of Croatia on my mobile for example.

For smaller areas I can scroll manually in Vespucci and count on it caching
it (for example if it is for evening walk); but for a week's bicycle ride
it's not really feasible offline without lot of preparation (and probably
not even then)...

>>       - require manually pointing at exact location for each address node
> This is naturally only an issue if you are adding nodes and not adding 
> address tags to a building outline, but as nearly all such problems the 

Correct, for nodes.  If building was correctly marked as outline, I would
probably be using StreetComplete exclusively and not even considering other
tools, due to both simplicity and quickness of use and no postprocessing.
(is is only when buildings are not fully mapped, that StreetComplete
becomes useless at the address mapping)

> solution is to zoom in a bit more (I'm a bit puzzled by the "exact" 
> location bit though).

To map address node in Vespucci where there is no building outline
pre-mapped, I need to (imagine building number "42" on my left side
as I'm walking down the street):

- determine in what direction I am moving on the map (having current GPS track enabled and shown helps a lot)
- determine what location is on left side of my current position (depends on above observation, of course)
- if having aerial imagery, find the building, possibly needing zooming in/out (if no imagery, then just click few meters off the road in correct direction)
- long press on the location where I want to add the node
- click on address icon
- (unless AI has done the job automatically), click on housenumber field and type a number like "42"
- click on OK icon

Compared to Keypad Mapper 3 UI:
- type 42
- press left arrow

So, even when AI in Vespucci works flawlessly, KP3 is usually still faster for me.  
And when housenumber needs to be enter manually in Vespucci, KP3 wins hands down.
(Also less things to go wrong in KP3 due to human error, as less human actions are required)

Sure, KP3 needs some postprocessing in JOSM for ideal results, but even if
you just upload those address nodes which are put on left/right side of the
street without any extra work in JOSM, it is still great and immediately
useful address collection!

And if I DO decide to do postprocessing in JOSM (for example, to draw building
outlines which were not drawn), JOSM works much faster for me than Vespucci:
bigger screen, more precise mouse movements than my frozen fingers in gloves
on small mobile screen, fast/free internet access, and JOSM "building tools"
plugin makes drawing building an one-click breeze (it also autoconverts 
address data from node to building outline, makes building outlines parallel
automatically when wanted etc.)


>>       - classic full Android keyboard is small and unoptimized for this specific use
>
> Sure, but if somebody actually -wants- a solution to this and is willing
> to test and provide feedback this can naturally be added, see the open 
> issue I referred to.

I've added this proposal to:
https://github.com/MarcusWolschon/osmeditor4android/issues/574

I would like to have simple UI for addresses like KP3:
https://wiki.openstreetmap.org/w/images/9/9b/Keypadmapper2_1.8_keypad.png

(note: apart from quick and big keyboard shown on picture, clicking on address
field opens up standard android keyboard, to deal with i18n and other chars.
But this keypad alone gets used more in than 95% of my cases)

However, I do not program for Android, so unfortunately I cannot contribute
PR for the feature, but would be grateful if someone else did it (hint, hint).  
I can test beta .APKs (provided it does not require google play, which I don't 
have) and provide feedback on US and HR keyboards.

-- 
Opinions above are GNU-copylefted.




More information about the talk mailing list