Ideas for mapping/locating sidewalks within the Project Sidewalk tool ecosystem

[misaugstad]

As we have talked about many times in meetings and conversations, and also in a recent issue #1126... Placing a no sidewalk every panorama or every few panoramas is very inefficient; lack of sidewalk is (almost) never really a "point" problem, it is inherently something that stretches across many panoramas. This also makes it difficult to evaluate the accuracy of no sidewalk labels that are placed.

So I want to have this thread as a place where we can talk about ideas that have come up for changes to the tool (or new tools) that would help with labeling no sidewalk.

Some ideas:

1. Giving the user the ability to get a larger top-down view of the sidewalk, ideally with satellite imagery. They could then do something like click on the start and end points of missing sidewalk from that top-down map. Or clicking multiple points to create a sequence of connected line segments so that arbitrary geometries can be taken into account. One way this could be incorporated into the tool, is possibly at the end of every street segment that you audit, this top-down satellite image pops up, with the street you just audited highlighted, and you mark this on the street segment you just did, then you go back to the normal tool to audit the next street.
2. Having something similar to polygons in Tohme, except that it spans multiple panos, is a line and not a polygon, and is visualized on the top-down map instead of SV image. So you can imagine seeing where the sidewalk ends, placing a no sidewalk label, walking for awhile, then when you place another no sidewalk label, it draws a line on the top-down map b/w those two labels. You could then continue taking a few steps and placing another no sidewalk label, and it would connect this to the previous label, making a sequence of connected line segments. Then when the sidewalk starts back up, you place your final no sidewalk label, and somehow mark it as an endpoint.

[jonfroehlich]

I like the first one; however, note that @kotarohara and I brainstormed and mocked up some of these interfaces early on (I think with a ugrad student) and one thing that we noticed is that sidewalks are not always visible in satellite imagery due to building shadows (a problem downtown) or tree/vegetation coverage (a problem more in less urban areas)--but both are fairly common. Perhaps could get around this by showing a birds eye view (which Microsoft Bing maps offers for example) or satellite+GSV (though not sure how to do latter in an elegant manner)?

[jonfroehlich]

Coming back to this because mapping/locating sidewalks (and building the underlying sidewalk topology) is a critical part of surveying a community's sidewalk infrastructure.

There is a recent journal paper published in Feb 19, 2021 entitled Sidewalk extraction using aerial and street view images, which uses deep learning to piece together sidewalks by combining both aerial and street view imagery.

Ning H, Ye X, Chen Z, Liu T, Cao T. Sidewalk extraction using aerial and street view images. Environment and Planning B: Urban Analytics and City Science. February 2021. doi:10.1177/2399808321995817

[misaugstad]

@jonfroehlich just shared this interesting dataset on Slack! https://www.nearmap.com/us/en/current-aerial-maps-coverage

[jonfroehlich]

And Maryam shared this super high resolution (3" res) aerial photography with CC by 4.0 license of Washington DC: https://opendata.dc.gov/datasets/2cb9fe0f4b444ebbae359b8889f39dda.

Could be great for our disability parking research.

Or for labeling other properties of sidewalks (including sidewalk mapping)