Yu-Min Lin, Albert Huynh, Andrew Lanckriet, Gert Barrington, Luke Overall System: <p><i>Left</i> - Participants are provided sub-sectioned imagery tiles in random order to label features such as roads, rivers, and ancient structures through an online interface, we have shown three semi-overlapping tile croppings. <i>Center</i> - A geospatial map of observed ground features is generated though the combined input of tens of thousands of independent inputs. Overlapping regions in image tile croppings facilitate a global KDE for a universal comparison of saliency across the entire data set. Regions of highest KDE likelihood (based on user inputs) are highlighted. <i>Right</i> - Aerial photography and ground exploration of the location identified by crowd reveals a circular “khirigsuur” burial mound with Bronze Age <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0114046#pone.0114046-Allard1" target="_blank">[24]</a>, <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0114046#pone.0114046-JacobsonTepfer1" target="_blank">[25]</a> characteristics. Satellite imagery provided courtesy of the GeoEye Foundation.</p> training mechanism;consensus;scale survey;Genghis Khan;10 K;kernel density estimation;networked groups;anomaly detection;feature categorizations;landscape;peer feedback loop;Genghis Khan Massively;participant;reasoning engine;knowledge generation;National Geographic expedition 2014-12-30
    https://plos.figshare.com/articles/figure/_Overall_System_/1281724
10.1371/journal.pone.0114046.g002