2018 ESA Annual Meeting (August 5 -- 10)

COS 18-6 - A new workflow to remove shadow from UAV imagery with a view to improve vegetation structure delineation in a classical bog

Monday, August 6, 2018: 3:20 PM
353, New Orleans Ernest N. Morial Convention Center
Mustafizur Rahman, Geography, University of Calgary, Canada, Greg Mcdermid, Geography, University of Calgary and Maria Strack, Geography and Environmental Management, University of Waterloo, Canada
Background/Question/Methods: Abundance, density and distribution of different types of vegetation within a wetland (i.e., a classical bog in our case) ecosystem provide import information on the state of the ecosystem and the carbon balance. High spatial resolution UAV data can be very useful to obtain orthophoto and dense point cloud that are appropriate for vegetation structure mapping. However, the presence of shadows from vegetation, terrain, and other elevated features represent lost and/or impaired data values that hinder the quality of optical images acquired under all but the most diffuse illumination conditions. Fortunately, the flexibility and low cost of re-deployment of the platform also presents opportunities, which we capitalize on in a new workflow designed to eliminate shadows from UAV-based orthomosaics. Our three-step procedure relies on images acquired from two different UAV flights, where illumination conditions produce diverging shadow orientations: one before solar noon and another after. From this multi-temporal image stack, we first identify and then eliminate shadows from individual orthophoto components, then construct the final orthomosaic using a feature-matching strategy with the commercial software package Photoscan.

Results/Conclusions: We demonstrated our workflow in a study area located approximately 40 km north of Peace River in the Canadian province of Alberta, Canada: a complex treed bog containing wide variety of shadows. To evaluate the performance of our workflow, we compared a shadow-reduced orthomosaic to a traditional one obtained from a single flight. We classified the raw orthomosaic into two classes – shadow and other – using an unsupervised (K-Means) decision rule. According to our classification, 22.3% of the study area was covered by shadow in the traditional orthomosaic, for which it was not possible to map the surface/vegetation. On the other hand, the shadow-reduced orthomosaic was able to remove shadows from these areas and thereby reveal the underlying vegetation structure.