Generation of DSM of forest crown generated by vertical + oblique stereo pair images taken by small-sized UAV

Abstract

Photographic surveying using a small-sized UAV (Unmanned Aerial Vehicle) has recently attracted attention. The SfM (Structure from Motion) method makes it possible to create 3D point clouds and a 3D model from multiple 2D images. Furthermore, an orthomosaic photograph and DSM (Digital Surface Model) can be generated from the 3D model. It has been reported that the precision of the point clouds becomes low when the target is vegetation due to insufficient resolution of images, the vegetation moving in the wind, and shadow areas in the images. This study, therefore, created a DSM of a forest crown using nadir+oblique stereo pair images taken by a small-sized UAV. The study was carried out in the larch forests at the foot of the Yatsugatake Mountains, Yamanashi Prefecture, Japan, in July, 2015. A UAV with a digital camera flew over the study site to acquire crown images in the nadir and oblique directions using an autopilot system. We first generated dense point clouds, from which we then generated orthomosaic photographs and DSMs following three patterns: (1) 70 nadir images taken at an altitude of 100m above the ground; (2) (1) plus 54 nadir images taken at an altitude of 50m above the ground; and (3) (1) plus 54 oblique images taken at an altitude of 50m above the ground. Under Pattern (1), 17.5% of the total area had no point clouds, while Patterns (2) and (3) showed 12.8% and 9.7%, respectively, with no point clouds. We obtained DSMs with a spatial resolution of 2.0∼2.5 cm for all three patterns. Some areas of the DSM of Pattern (1) showed less surface roughness; such areas decreased in Patterns (2) and (3). In conclusion, the present study demonstrates an improvement in the reproducibility of DSMs by adding oblique images in comparison with the use of nadir images alone.