Label Types in Project Sidewalk: Crosswalks, Pedestrian Signals, and Beyond

[jonfroehlich]

From user feedback:

Could have:

• A crosswalk label for crosswalks
• A 'missing crosswalk' label to indicate the need for a crosswalk

[kotarohara]

We had a discussion about this. I don't think crosswalk should be included as a part of the current interface, which is focused on collecting data for people with mobility impairments.

But I agree it would be nice to collect crosswalk data along with all the other urban data.

[jonfroehlich]

I agree with you. What should we do with feature requests like this? Close them out or keep them open so that we remember them...

[misaugstad]

Just putting in my 2 cents as I come across this: I would think this is more easily solved using CV and satellite imagery, and isn't a great fit for our tool.

[jonfroehlich]

Why not a fit for our tool?

Sent from my iPhone

On Jun 19, 2018, at 7:29 AM, Mikey Saugstad notifications@github.com wrote:

Just putting in my 2 cents as I come across this: I would think this is more easily solved using CV and satellite imagery, and isn't a great fit for our tool.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[misaugstad]

Just because it seems like a CV application that wouldn't need much user input. I suppose I may have been taking a slightly more narrow view of the project than I should have. We could easily have a sub-part of Project Sidewalk that does this analysis of satellite imagery and have user validation missions that are used to train the models that identify the crosswalks.

[jonfroehlich]

So, I would still advocate trying to include a crosswalk label. In my discussions with DOTs (including DDOT), crosswalks are always mentioned as important assets that need to be tracked and monitored because they wear out due to weathering (i.e., the paint fades away).

Roberto Manduchi and colleagues have done quite a bit of work on using Computer Vision to automatically recognize crosswalks; however, they've found that simply relying on satellite imagery is insufficient due to occlusion from shadows and trees (try this out yourself--load up downtown Washington DC and there are a surprising number of sidewalk features obscured by such things). So, at ASSETS'15, they demonstrated a technique of combining satellite imagery + GSV in their computer vision pipeline:

Dragan Ahmetovic, Roberto Manduchi, James M. Coughlan, and Sergio Mascetti. 2015. Zebra Crossing Spotter: Automatic Population of Spatial Databases for Increased Safety of Blind Travelers. In Proceedings of the 17th International ACM SIGACCESS Conference on Computers & Accessibility (ASSETS '15). ACM, New York, NY, USA, 251-258. DOI: https://doi.org/10.1145/2700648.2809847

There appears to be more recent work on this (haven't read it yet) by other research groups, including:

Koester, D., Lunt, B., & Stiefelhagen, R. (2016, July). Zebra crossing detection from aerial imagery across countries. In International Conference on Computers Helping People with Special Needs (pp. 27-34). Springer, Cham.

Berriel, Rodrigo F., et al. "Deep Learning-Based Large-Scale Automatic Satellite Crosswalk Classification." IEEE Geoscience and Remote Sensing Letters 14.9 (2017): 1513-1517.

Ahmetovic, D., Manduchi, R., Coughlan, J. M., & Mascetti, S. (2017). Mind your crossings: Mining GIS imagery for crosswalk localization. ACM Transactions on Accessible Computing (TACCESS), 9(4), 11.

Berriel, R. F., Rossi, F. S., de Souza, A. F., & Oliveira-Santos, T. (2017). Automatic large-scale data acquisition via crowdsourcing for crosswalk classification: A deep learning approach. Computers & Graphics, 68, 32-42.

It may also be, as you say, that we would incorporate labeling crosswalks as a separate mission type or mini-game within Project Sidewalk (that also uses top-down imagery).

[jonfroehlich]

I came here to say that I still think we should consider adding a crosswalk label button (maybe as one of the drop down 'Other' options). :)

[jonfroehlich]

Ha, was just going to add a new Issue about adding a Crosswalk Label type but then found this. I've been in discussions with some public health researchers interested in exploring pedestrian injuries and fatalities at intersections and one key thing that's helpful here is knowing whether there is a painted crosswalk.

Pinging @misaugstad so he sees this.

[jonfroehlich]

And like all other label types, users could rate severity and provide specific tags. Here are some examples:

[jonfroehlich]

Coming back to this again, from Shiloh Deitz's work at Univ. of Oregon (she also cites Lobben and Perdue in coming up with this list but I couldn't find underlying reference).

These are accessibility barriers or facilitators that we may want to consider adding either as label types or as tags in future (some we can grab from available data sources, like speed limits):

[jonfroehlich]

This is the type of categories the Urban Accessibility team in Amsterdam is thinking about and examining:

[jonfroehlich]

In Dr. Eisenberg's audit study, they also assessed pedestrian signals and crosswalks. I've attached their codebook, which may help inform our buttons + tags: AuditTool.xlsx

[jonfroehlich]

As we're now working on adding a crosswalk label, we need to start discussing tags.

From the previously referenced AuditTool.xlsx from Yochai's group, they have the following "codes" for crosswalks, which we could convert to tags:

Also, another one that Amy shared over email was "Is the crosswalk aligned with the curb ramp (but this may already be captured sufficiently in the curb ramp label)"

I think all four of these sound good as a starting point (will have to make them shorter to use them as tags).

[jonfroehlich]

For Professor Quistberg's work in Colombia, they are examining:

[jonfroehlich]

We now have crosswalks and pedestrian signals! Closing this for now :)