This will count as one of your downloads.
You will have access to both the presentation and article (if available).
Treetop detection and tree crown delineation are common tasks in forest-related studies since they are necessary steps for data analysis on an individual tree level. In recent years, related studies have concentrated on Machine Learning approaches for which training data are needed. The results of the methodology presented in this paper and applied to a freely available data basis can be used to train such Machine Learning algorithms.
The National Ecological Observatory Network (NEON) provides data products from 81 American field sites. This study evaluates the suitability of the NEON dataset – especially the Canopy Height Model (CHM) – for the automatic treetop detection and tree crown delineation in natural mixed forests that provide difficult real-world conditions. Both tasks are conducted exemplarily on two NEON field sites, BART and HARV. For comparison, data from a study area located in Meppen, Germany, is used.
The general workflow consists of three steps. First, the data is pre-processed by masking irrelevant pixels. Then, the treetops are detected with a method based on the Top-Hat by Reconstruction operation. Finally, the tree crowns are delineated with a region-growing segmentation method proposed by Dalponte and Coomes (2016). Both methods rely on tree height information to locate the treetops and extract the tree crown boundaries. Achieved results reveal that the NEON CHM is suitable for treetop detection. However, the CHM’s spatial resolution is too coarse for tree crown delineation, i.e. further data has to be considered for an accurate outcome.
In this paper, the procedure regarding point cloud generation of urban scenes, with images from the nadir RGB camera, is described in detail. To produce dense point clouds three main steps are necessary: generation of disparity maps, creation of depth maps, and calculation of world coordinates (X, Y, and Z).
To create disparity maps, two adjacent images (stereopair) were rectified. Afterwards, the PatchMatch Stereo (PMS) algorithm for 3D reconstruction was executed, since it is easy to implement and provides good results according to the Middlebury Computer Vision dataset. Some steps were parallelized to optimize execution speed. Since depth is inversely proportional to disparity, depth maps were calculated from disparity maps. The height of scene elements Z was obtained by subtracting their depth to the camera height.
To calculate the remaining world coordinates X and Y, the back-projection equation and the camera intrinsic and extrinsic parameters were used. To validate the PMS algorithm, its resulting point cloud was compared with a LiDAR point cloud and a PhotoScan point cloud. The root mean square errors of both comparisons showed similar values.
To achieve real-time processing, the incoming data streams are evaluated in small patches. In addition, the calculation time per patch must be lower than the recording/sampling time to ensure a real-time processing. Data meshing and projection of the images onto the mesh cannot be implemented in real-time using an off-the-shelf CPU. However, most of these steps are highly vectorizable (e.g., the projection of each LiDAR point into the camera images). In fact, modern graphics cards are highly specialized in computing such data types. Therefore, all computationally intensive steps were performed in the graphics card. Most of the steps for the terrain model generation have been implemented in CUDA and OpenCL. We compare both technologies regarding calculation times and memory management. The fastest technology was selected for each calculation step. Since the model generation is faster than the data acquisition time, the implemented software is real-time.
Our approach has been embedded and tested in a real-time system consisting of a modern reconnaissance system connected to a ground control station via a radio link. During a flight, a human operator in the ground control station is able to observe a texturized terrain model, which was recently generated. The user is able to zoom in an interesting area.
Expanding the concept of our preceding remote sensing platform developed together with OHB System AG and Geosystems GmbH, in this paper we present an airborne multi-sensor system based on a motor glider equipped with two wing pods; one carries the sensors, whereas the second pod downlinks sensor data to a connected ground control station by using the Aerial Reconnaissance Data System of OHB. An uplink is created to receive remote commands from the manned mobile ground control station, which on its part processes and evaluates incoming sensor data. The system allows the integration of efficient image processing and machine learning algorithms.
In this work, we introduce a near real-time approach for the acquisition of a texturized 3D data model with the help of an airborne laser scanner and four high-resolution multi-spectral (RGB, near-infrared) cameras. Image sequences from nadir and off-nadir cameras permit to generate dense point clouds and to texturize also facades of buildings. The ground control station distributes processed 3D data over a linked geoinformation system with web capabilities to off-site decision-makers. As the accurate acquisition of sensor data requires boresight calibrated sensors, we additionally examine the first steps of a camera calibration workflow.
Concept and integration of an on-line quasi-operational airborne hyperspectral remote sensing system
View contact details
No SPIE Account? Create one
