submission voting
voting is closed.


Automatic Georeferencing

short description

Use Computer Vision to extract features from vector & raster data, match features using Artificial Intelligence, & get geotransform params.




Data Integration


Automatically combine vector and raster geo-datasets.


The Smart M.App enhances the value of disparate datasets through integration and because through automation allows large sets of raster data to be pulled into a vector based mapping environment.

All local governments in the US use Assessor Maps, which are comprised of a set of tiled maps organized by book and page. Many Surveyor departments in the US maintain CAD/GIS systems. These two mapping systems are used for distinct purposes so the maps of the same geography may be similar, but have their unique differences. For example, the Assessor often subdivides land to so he can tax different parts differently or may aggregate adjacent parcels so he can generate a single bill, whereas the surveyors are intered only in the lines of ownership *as defined* by recorded maps (which means their maps don't necessarily reflect the real world). Such a Smart M.App would allow these datasets to be integrated providing greater utility for the GIS community.

comments (public)

  • Showbox Apk Dec. 27, 2017, 1:47 a.m. PST
    The US use Assessor Maps, which are comprised of a set of tiled maps organized by book and page. Many Surveyor departments in the US maintain CAD/GIS systems. Thanks for sharing post it's very informative. https://crumbles.co/best-kodi-add-ons/
  • Eugene Levin May 10, 2016, 4:44 p.m. PDT
    As a Surveying Engineering program chair I was not able to vote for another App, including one that was submitted by me and my students :)
    • Noel Khan June 23, 2016, 12:40 p.m. PDT

      Thank you for the support!

      I've worked on this problem on and off since 2006 and built a batch oriented system a couple years ago that uses OpenCV and AI optimization techniques. There's a reference to an IEEE paper from this past December on my LinkedIn page that describes the problem and solution if you're interested. The current batch system solves 1000 of these problems a day on my 12-node cluster (~10min/problem), but involves 1 day of overhead to configure. To put the throughput into perspective, the County of Orange Public Works Dept has 3 people on payroll who do 1000 maps a year manually.

      One of the reasons porting this system to Smart M.Apps is attractive is because it allows self service for smaller data volumes so my time is soaked up with pilots and folks can self serve. Also, I don't have to spin up additional instances on EC2 since this platform is already there and inherently elastic (based on your subscription level).

      For the purposes of this 2 month contest, the scope is going to be limited to assessor maps and parcel layers, which is data County level governments maintain. There are 3,000 counties and I estimate 15M such maps. Although many counties may not have a parcel layer, Core Logic maintains a national parcel layer based on survey plans. Also note that I've developed an alternative solution to this problem that may reduce the computational time to a negligible amount, so I'm pretty exciting about putting this process online.

      The road map includes generalizing the system so it can work with arbitrary content, but also different permutations of data types (e.g., georef a video to a model, register multiple rasters (for military use)). The unicorn is extending the system to 3D.

      The primary risk and challenge porting this to Smart M.Apps is whether there are sufficient baked-in capabilities to allow this type of processing or API to build something to fill gaps.

      Thanks again for the vote!