Road Network and Travel Time Extraction from Multiple Look Angles with SpaceNet Data

  • 2020-01-16 16:23:59
  • Adam Van Etten, Jacob Shermeyer, Daniel Hogan, Nicholas Weir, Ryan Lewis
  • 6

Abstract

Identification of road networks and optimal routes directly from remotesensing is of critical importance to a broad array of humanitarian andcommercial applications. Yet while identification of road pixels has beenattempted before, estimation of route travel times from overhead imageryremains a novel problem, particularly for off-nadir overhead imagery. To thisend, we extract road networks with travel time estimates from the SpaceNet MVOIdataset. Utilizing the CRESIv2 framework, we demonstrate the ability to extractroad networks in various observation angles and quantify performance at 27unique nadir angles with the graph-theoretic APLS_length and APLS_time metrics.A minimal gap of 0.03 between APLS_length and APLS_time scores indicates thatour approach yields speed limits and travel times with very high fidelity. Wealso explore the utility of incorporating all available angles during modeltraining, and find a peak score of APLS_time = 0.56. The combined modelexhibits greatly improved robustness over angle-specific models, despite thevery different appearance of road networks at extremely oblique off-nadirangles versus images captured from directly overhead.

 

Quick Read (beta)

loading the full paper ...