![]() ![]() Bottom left: Zoomed in region around the green midpoint. Top: Example road edge (yellow), with midpoints of each segment shown in blue. In the multi-class case, if the majority of the the high confidence pixels in the prediction mask patch belong to channel 3 (corresponding to 31-40 mph), we would assign the speed at that patch to be 35 mph.įor the continuous case the inferred speed is simply directly proportional to the mean pixel value.įigure 8: Segment speed estimation. ![]() The speed of the patch is estimated by filtering out low probability values (likely background), and averaging the remaining pixels. For each segment in the edge, at the location of the midpoint of the segment we extract a small 8 × 8 pixel patch from the prediction mask. This is accomplished by analyzing the prediction mask at the location of the segment midpoints. Accordingly, we attempt to estimate the speed of each segment in order to determine the mean speed of the edge. The majority of edges in the graph are composed of multiple segments (e.g. We estimate travel time for a given road edge by leveraging the speed information encapsulated in the prediction mask. We also follow and remove terminal vertices that lie on an edge less than 10 pixels in length,Īnd connect terminal vertices if the distance to the nearest non-connected node is less than 20 pixels.įigure 7: Graph extraction procedure. We remove disconnected subgraphs with an integrated path length of less than To close small gaps and remove spurious connections not already corrected by the opening and closing procedures, The graph created by this process contains length information for each edge, but no other metadata. This skeleton is rendered into a graph structure with a modified version of This binary mask is then refined using techniques such as opening and closing, from which We begin by smoothing and flattening the final output mask to create a binary prediction mask. The output of the segmentation mask step detailed above is subsequently refined into road vectors. Right: OSM road labels (orange) and SpaceNet building footprints (yellow) in some cases road labels are misaligned and pass through buildings. Left: OSM roads (orange) overlaid on Khartoum imagery the road traveling left to right across the image is missing. Figure 1: Potential issues with OSM data. For example, following Hurricane Maria, it took the Humanitarian OpenStreetMap Team (HOT) over two months to fully map Puerto Rico. An active community works hark to keep the road network up to date, but such tasks can be challenging and time consuming in the face of large scale disasters. Yet, in developing nations OSM labels are often missing metadata tags (such as speed limit or number of lanes), or are poorly registered with overhead imagery (i.e., labels are offset from the coordinate system of the imagery), see Figure 1. For example, in many regions of the world OpenStreetMap (OSM) road networks are remarkably complete. Indicating that optimizing routing for travel time is feasible with this approach.Įxisting data collection methods such as manual road labeling or aggregation of mobile GPS tracks are currently insufficient to properly capture either underserved regions (due to infrequent data collection), or the dynamic changes inherent to road networks in rapidly changing environments. Metric scores decrease by only 4% on large graphs when using travel time rather than geometric distance for edge weights, We also test our algorithm on Google satellite imagery with OpenStreetMap labels, and find a 23% improvement over previous work. We compare SpaceNet labels to OpenStreetMap (OSM) labels, and find that models both trained and tested on SpaceNet labels outperform OSM labels by ≥ 60 %.įor a diverse test set of SpaceNet data and a traditional edge weight of geometric distance, we find an aggregate of 5% improvement over existing methods. Including estimates for travel time permits true optimal routing, not just the shortest geographic distance. We call this approach City-Scale Road Extraction from Satellite Imagery v2 (CRESIv2). To this end, we explore road network extraction at scale with inference of semantic features of the graph, identifying speed limits and route travel times for each roadway. Automated road network extraction from remote sensing imagery remains a significant challenge despite its importance in a broad array of applications. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |