An investigation of tree extraction from UAV-based photogrammetric dense point cloud

Manually, tree detection with terrestrial field work is a nonprofit labor in terms of time, cost, and manpower. As a rapid alternative, terrestrial and airborne laser scanners are widely in use for data collecting. But this active remote sensing technology is expensive, especially in local small areas. At this point, unmanned aerial vehicles stand as a new opportunity for data collection platforms for tree detection in both large and local study areas. This study shows the usage of unmanned aerial vehicle as a platform to collect aerial images. The images are used to generate 3D dense point clouds. This point cloud is investigated with the random sample consensus algorithm in order to detect tree. The trees are assumed as or in a cylinder which is geometrically defined with respect to tree’s parameters such as radius. According to the results, the RANSAC algorithm is successful in the detection of trees from unclassified image-based dense point cloud. The 232 individual trees have managed to extract with a rate of 70.1% from 3 different study sites.