[Imports] Uploading sidewalks in San Jose, California, US

[Minh Nguyen]

On 15/10/2017 15:08, Michael Reichert wrote:
> Hi Vivek et.al.,
> 
> Am 20.09.2017 um 09:20 schrieb Frederik Ramm:
>> On 19.09.2017 21:09, Vivek Bansal wrote:
>>> The Wiki:
>>> https://wiki.openstreetmap.org/wiki/Santa_Clara_County,_California/San_Jose_Sidewalk_Import
>>
>> Let me try to understand this better. The wiki page says
>>
>> "We recognize we could be creating sidewalk islands since we aren't
>> dealing with crossings or connecting the sidewalks to streets but most
>> routing software doesn't use sidewalks and typically uses left/right
>> tags. Regardless, adding sidewalks will eventually make things better."
>>
>> At the same time you are talking about the "pedestrian network" and
>> "fundamental infrastructure mapping". But it isn't a network if it isn't
>> connected, is it? You must be mapping zebra (and other) crossings too,
>> so that actual pedestrian routes can be computed, or else the use of the
>> data you're adding would be limited to walking between places on the
>> same block of houses?
>>
>> Please confirm that your process will not be limited to "uploading"
>> sidewalks but also ensuring they are properly connected to what's there
>> on OSM already, and thereby (through crossings etc.), also to each
>> other. Otherwise what you are doing is nothing better than MS Paint for
>> maps - or potentially worse, since a pedestrian routing engine that
>> *today*, because of the lack of sidewalks, snaps start and destination
>> to the nearest road and leads the pedestrian along that, might choose to
>> snap to a sidewalk instead and be caught on a routing island.
> 
> I seems that you did not understand correctly what Frederik wanted to
> tell you.
> 
> If you upload sidewalks as separate ways, they must be connected with
> the existing network in the same changeset which uploads them.
> Unfortunately, the data you uploaded already is not connected to the
> remaining network at all. Please fix these thousands of routing islands
> and dead-ends within the next days or revert your import. Please pause
> your import and get more familiar with OSM before you continue.

Hi, I'm one of the primary mappers in the affected area, having mapped 
42% of the sidewalks in San Jose by hand before the import started. [1] 
I've been working with Vivek to ensure that the sidewalk import is part 
of a broader, long-term effort to improve OSM's quality in the area, 
through both imports and manual edits. Most of my hand-mapped sidewalks 
will be deleted in the conflation process, but nonetheless I welcome 
this import because it'll give hand-mappers like me (what will be) 
high-quality data to build upon.

Having the sidewalks in first -- even if disconnected from the road 
network -- makes it easier to fill in the crosswalks. The main challenge 
mapping crosswalks in some neighborhoods is that tree cover makes it 
difficult to determine where the crosswalks meet the sidewalks -- you 
can see the crosswalks but not the street corners. But if reasonably 
accurate sidewalks are present (as they will be with this import), that 
becomes much more straightforward.

So far, we've learned a lot in terms of import logistics, and we were 
hoping to have each contributor focus and get the hang of this one 
import before branching out with an additional project to fill in the 
gaps. It would be a shame and pretty disruptive if the sidewalks were to 
be reverted at this point, since we are committed to adding the crosswalks.

> OpenStreetMap intends to be a routeable data set. Lots of users fetch
> data updates for routing every day, every week or every month and rely
> on us. If you upload broken data, their routing software will return low
> quality results until the next update.
> Example of a result which harms the reputation of OSM:
> http://www.openstreetmap.org/directions?engine=graphhopper_foot&route=37.31370%2C-121.75710%3B37.31290%2C-121.75690

That is indeed a problem we hope to eradicate once the crosswalk mapping 
project gets underway. It isn't unique to imports, though: for example, 
the adjacent town of Los Gatos has been mapped fairly extensively with 
sidewalks, and many crossings are missing there too:

http://www.openstreetmap.org/directions?engine=graphhopper_foot&route=37.22855%2C-121.97027%3B37.22866%2C-121.97025

http://www.openstreetmap.org/directions?engine=graphhopper_foot&route=37.22914%2C-121.96464%3B37.22918%2C-121.96475

Even where crossings were added, they weren't connected to the road 
network, leading to very roundabout routing:

http://www.openstreetmap.org/directions?engine=graphhopper_foot&route=37.22733%2C-121.96631%3B37.22697%2C-121.96609

These problems come up often from mappers who are unfamiliar with 
pedestrian router needs. I mention these examples not to pass the blame 
but rather to point out that the problem is not new in this area, yet 
we're working deliberately to ensure that mappers won't have to worry as 
much about these issues in the future.

[1] http://overpass-turbo.eu/s/smw versus http://overpass-turbo.eu/s/smv

-- 
minh at nguyen.cincinnati.oh.us