We present an open access dataset for development, evaluation, and comparison of algorithms for individual tree detection in dense mixed forests. The dataset consists of a detailed field inventory and overlapping UAV LiDAR and RGB orthophoto, which make it possible to develop algorithms that fuse multimodal data to improve detection results. Along with the dataset, we describe and implement a basic local maxima filtering baseline and an algorithm for automatically matching detection results to the ground truth trees for detection algorithm evaluation.
From a paper by I. Dubrovin et al in Nature.
Forests are an essential renewable natural resource and an important dynamic part of the global carbon cycle. Responsible forest management allows for efficient use of it as a resource and regulation of atmospheric but requires up-to-date data about forest attributes such as distribution of species, above-ground biomass, age and height of the trees, and others1,2.
This drives the need for the ability to quickly and accurately map forest attributes, often covering incredibly large areas, which makes manual forest inventories too costly and time-consuming for extensive, repeated monitoring.
LiDAR has been used to augment and extrapolate limited field measurements to larger areas for a long time3. It is an active sensor, which means it does not depend on environmental conditions such as lighting, which greatly affects passive sensors like cameras and greatly reduces the universality of methods developed for them. It also penetrates the canopy and provides information about the vertical structure of the forest, including the shape of the terrain and lower levels of trees.
UAV LiDAR, out of all other ways to collect LiDAR data, such as from planes or from the ground, is relatively accessible and allows for most configuration of attribution parameters, with a good balance of data quality, cost, and effort.
The most common way to use LiDAR for mapping forests in industry is the area-based approach. It consists of calculating aggregated point cloud metrics for available ground plots, fitting statistical models to predict required attributes, and applying these models on wall-to-wall metrics to extrapolate the measurements4. Area-based approach can work with low density LiDAR point clouds, is easy to understand, implement, and extend with other data sources, but the spatial resolution of the results is very coarse. In contrast, modern sensors allow for much more detailed measurements which make it possible to not aggregate at all, instead working on the scale of individual trees.
For the complete article open access dataset for UAVs CLICK HERE.
Note – If you have a lidar related news story that you would like us to promote, please forward to editor@lidarnews.com and if you would like to join the Younger Geospatial Professional movement click here
