engleski

Testing the UAV-based Point Clouds of Different Densities for Tree- and Plot-level Forest Measurements

In the last 10 years, photogrammetric methods have gained a lot of attention in the forest inventory research and applications. This is caused by several factors. The first factor is advancement in the Computer Vision (CV) algorithms that neglected the usually complex and time-consuming photogrammetric workflow. The second factor is an advancement in sensor technology that consequently made the UAVs (photogrammetric platforms) more affordable. Due to the flexibility of use, price and accuracy of the products, UAV based photogrammetry has recently attracted great attention in many environmental applications, including forest inventory. Algorithms that brought photogrammetry to applicable level in forest inventory are Dense Image Matching (DIM) algorithms. They are allowing the user to produce a high- density point cloud (match each pixel) that represents the forest canopy and that can be used for further production of the Digital Surface Model (DSM) and the forest parameter estimation in combination with precise Digital Terrain Model (DTM). However, this is the most time-consuming process in the photogrammetric workflow, therefore automatically downsampled images are usually used in this process. This study aims to test point clouds of the different densities (produced with the different pyramid level of input images). Up to four pyramid levels of the images were used to produce point clouds by utilizing the algorithm implemented in the Agisoft PhotoScan software. The first test was conducted with the full resolution images (Level 0), the second with 1/2 (Level 1), the third with 1/4 (Level 2), the fourth with 1/8 (Level 3), and the last test was conducted with the 1/16 of the image resolution (Level 4). The study was conducted in the lowland pedunculate oak (Quercus robur L.) forests of Pokupsko Basin (Central Croatia). Point clouds were generated for the 54 circular plots with the radius ranging from 8 to 20 m. The UAV imaging was done using Trimble UX5 HP with Sony Alpha ILCE-R7 on 30 and 31 May 2017. Finally, 1441 images with the Ground Sample Distance (GSD) of approximately 8 cm were acquired. A high end- lap of 90% and side-lap of 80% were defined by the flight plan. UAV-based point clouds of different densities were tested on the tree- and plot-level using field reference data. Namely, for both tree- and plot-level, tree heights obtained from point clouds were compared with field-measured tree heights. Additionally, UAV-based point clouds were compared with the available Light Detection and Ranging (LIDAR) point cloud at the selected vertical profile of the 430 m length throughout the forest stand. The results of the conducted tests suggest that for the metric information extraction, the highest pyramid level and the most time-consuming process is not justified to use since benefits of such process are insignificant.

Unmanned Aerial Vehicle ; point cloud ; image pyramid ; tree height ; forest inventory

nije evidentirano