A 1.478Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE
The case of "Comparing Photogrammetric and Airborne-LiDAR Point Clouds in Individual Tree Crown Delineation and Species Identification" was presented in the 2015 Silvilaser conference in France.
Project Info
Organization Name: Department of Geography, University of Quebec at Montreal, Canada
Country: Canada
Industry: Natural Resources and Management
Project Date: March, 2015
Project Managers/Partners: Benot St-Onge, Flix Antoine Audet
Project Size: 133.6 km2
Number of Images: 376 images
GSD: 10 cm
The main purpose of the project was to interpret the delineation of individual trees and to identify the tree species by creating a hybrid canopy model (CHMs). Individual Tree Crowns were generated from both the photogrammetric and LiDAR point clouds, using an in-house algorithm that relies on height-adaptive Gaussian filtering, local maxima detection, edge detection, and region growing around the local maxima. The algorithm is controlled by stopping conditions based on edges, tree size and shape.
The images were taken from a metric camera- Vexcel (now Microsoft) UltraCAM-XP, onboard a fixed wing aircraft. The GSD of the project was 10 cm. Images where acquired using a standard flight plan for air photo acquisition, i.e. along parallel flight lines laid out so that a continuous block of land can be entirely covered. All photos were vertical (nadir), and 245 ground control points were used for the full area.
The images were calibrated using Pix4Dmapper. In the point densification process in the software, we chose the option 2 and 3 for minimum matches. It turned out that to densify point cloud with minimum 2 matches (more points but with slightly less accuracy) achieved a better result after interpolation because the minimum 3 matches contains less points and thus more null values when comparing to the LiDAR point clouds.
The quality of the photogrammetric point cloud is such that even individual trees with elongated shapes can be resolved. In comparisons with other image matching software, Pix4DMapper stood out as the one providing the highest 3D quality over forested scenes.
This allowed us to automatically delineate individual trees and extract features such as height and species type (conifer vs. deciduous). This is the first demonstration that such results can be obtained from photogrammetric point clouds.
For more information, visit: www.pix4d.com
A 1.478Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE