I'm currently using ArcGIS 10 Spatial Analyst's Cost Path tool to model pedestrian routes. I have sidewalks, fences, buildings and a ton of pedestrian scale features, all combined into a cost surface in which sidewalks are low cost, parks are slightly higher, crosswalks are still higher, and buildings, fences, and the like are barriers. The model works great to define optimal routes, compare routes to each other, assess the walkability of areas, and build walkingsheds.

But I have one major problem: bridges and tunnels. Because the cost surface forces all the features into a single 2D plane, I'm unable to model a bridge with a sidewalk that crosses over a street with a sidewalk. I must choose which one should get cut off. An alternative is to find all the bridges and tunnels in the City and model all combinations of under/over, but the model takes forever to run as it is! Is there a third party tool that can perform in least cost path in 3D, or am I missing a workaround of some kind?

  • Nice to see you here, Julie! You identify a fundamental limitation in performing costpath analyses with a purely gridded data structure. Hopefully respondents will suggest effective alternatives to consider.
    – whuber
    Nov 17 '11 at 22:19
  • 1
    Julie, have you ruled out vector solutions provided by the Network Analyst extension?
    – user681
    Nov 17 '11 at 22:52
  • Thank you both! there are two problems with a pure vector solution: (1) pedestrians move freely across areas, not on tracks like cars and trains e.g., crossing through parks and public areas; (2) I don't have the data to support a vector solution - my sidewalks and crosswalks are polygons; and generating good centerlines would be somewhat expensive (any ideas there?)
    – julie
    Nov 18 '11 at 12:17
  • 1
    @julie - WRT (1) As long as the peds are not trespassing, you might not want to build a model that encourages cross cutting even if it is the existing scenario. (2) There is prob. a way to grab centerlines from gaps in a polygon layer (new question?). Modelling wise, I would use regular road data and remove the edges with no sidewalks. I'll add trail data and manually trace whatever doesn't exist. Where roads and trails meet I'd create dummy nodes. The model would favours trails, roads, roads with no sidewalk, and then raster based data such that the route doesn't infringe anyone's property.
    – dassouki
    Nov 19 '11 at 21:09
  • @julie, out of curiosity: 1) what are the approximate dimensions of your project area, 2) what is the current grain/pixel size of your existing raster input, and 3) how much time actually elapses when you send it through the cost path tool? The more I think about this, the more I'm wanting these details.
    – elrobis
    Dec 2 '11 at 15:18

As @dassouki said, the Network Analyst solution could be suitable here, if you specify connectivity groups.

In your case the overpass and underpass would be in separate connectivity groups, so it would not be possible to traverse between them. (In cases where there are stairs, you can allow pedestrians to change elevations.)

In terms of allowing "unstructured" movement you could create vectors from your rasters using the FLOWACCUMULATION command, which would allow you to capture paths which don't follow traditional routes.

This wouldn't allow the same degree of freedom as the raster approach, but should solve the issue with the over/underpasses.

  • I agree and think it is a pure vector problem. The rasters should only be used to find routes that are not represented via vectors. Add to that most tunnels don't allow pedestrians.
    – dassouki
    Nov 28 '11 at 13:57
  • 1
    @dassouki Where I live, traffic is constrained by railroad rights-of way. At approximate one-kilometer intervals there are pedestrian underpasses to allow foot traffic. Sometimes these are shared with traffic, but most of them are dedicated (and they involve stairs). There is a similar circumstance involving the major highways surrounding the city: most of the crossings are underpasses and many admit foot traffic.
    – whuber
    Nov 28 '11 at 15:40
  • @whuber I guess I was referring to vehicle tunnels that go under a river, through a mountain or under a busy business / downtown core. Peds are not usually allowed in those high speed long facilities
    – dassouki
    Nov 28 '11 at 16:51
  • @whuber Julie could specify a cost/weight for the stairs, so a flat path would be chosen where it was faster, but using the stairs would also be possible. For disabled access the stairs could be classed as a barrier. help file entry Nov 28 '11 at 22:01
  • 1
    It doesn't always work like that, Stephen. For instance, within our fair city there is a walk along a river bank. Passing over that walk at high elevations are various highway overpasses, many of them with sidewalks. How do you create a raster that allows pedestrians to cross the river on these overpasses and also to get beneath the overpasses along the river walk, but not to connect directly from one to the other?
    – whuber
    Nov 28 '11 at 22:07

The question you have asked is non-trivial. I can't speak from experience on how ArcGIS handles least cost paths on a raster, because I haven't played around with it. But if you want to use a strictly raster approach, then you are correct in your assumption that you will need to model the over/under for bridges and tunnels.

The other major issue here is if you can modify ArcGIS's least cost path algorithm to use those over/under rasters when appropriate. This would require modifying the algorithm to look for certain features that will indicate that you need to load the appropriate over/under raster for traversing bridges or tunnels. My initial guess is that the ArcGIS least cost paths algorithm cannot be modified.

** DISCLAIMER ** My company has developed a solution for this problem and you can view a demonstration of it at http://www.groundguidance.com. The area is Lincoln Park in Chicago. This area highlights our multiplanar raster routing capabilites. Feel free to play around with it and if you have any questions feel free to ask.

  • 1
    Very interesting! Could you explain what you mean by a "multiplanar raster routing capability" or supply a reference to more information?
    – whuber
    Dec 2 '11 at 20:47
  • @whuber I don't know how much I an divulge here, as it is a proprietary solution. A general idea: given an extent, create a ground plane, and any other planes you need for that extent. The ground plane holds the majority of the raster data, while the extra planes contain "tunnels" that represent bridges, etc. When the algorithm reaches an entrance to these planar "tunnels", it is only allowed to pass through the "tunnel" to the other side. So there is no "suicide route" which jumps off a bridge to a path below it.
    – bp24
    Dec 2 '11 at 22:23

You might try to treat this as an indoor GIS problem. Implementing Dijkstra Shortest Route for 3D Indoor Navigation System

It's not exactly brain surgery - but it's close.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.