Heezin Lee

Imputation of individual Longleaf Pine (Pinus palustris Mill.) Tree attributes from field and LiDAR data

Abstract Light Detection and Ranging (LiDAR) has demonstrated potential for forest inventory at the individual-tree level. The aim in this study was to predict individual-tree height (Ht; m), basal area (BA; m2), and stem volume (V; m3) attributes, imputing Random Forest k-nearest neighbor (RF k-NN) and individual-tree-level-based metrics extracted from a LiDAR-derived canopy height model (CHM) in a longleaf pine (Pinus palustris Mill.) forest in southwestern Georgia, United States. We developed a new framework for modeling tree-level forest attributes that comprise 3 steps: (i) individual tree detection, crown delineation, and tree-level-based metrics computation from LiDAR-derived CHM; (ii) automatic matching of LiDAR-derived trees and field-based trees for a regression modeling step using a novel algorithm; and (iii) RF k-NN imputation modeling for estimating tree-level Ht, BA, and V and subsequent summarization of these metrics at the plot and stand levels. RMSDs for tree-level Ht, BA, and V were 2.96%, 58.62%, and 8.19%, respectively. Although BA estimation accuracy was poor because of the longleaf pine growth habitat, individual-tree locations, Ht, and V were estimated with high accuracy, especially in low-canopy-cover conditions. Future efforts based on the findings could help improve the estimation accuracy of individual-tree-level attributes such as BA.

Spark-based in-memory DEM creation from 3D LiDAR point clouds

ABSTRACT The significant performance improvement obtained by using Spark in-memory processing for iterative processes has led many researchers in various fields to implement their applications with Spark. In this study, we investigated the use of in-memory processing with Spark for creating a digital elevation model from massive light detection and ranging (LiDAR) point clouds, which can be considered an iterative process. We conducted our experiments on large high-density LiDAR data sets using two well-known interpolation methods: inverse distance weighting (IDW) and Kriging. Here, we designed our in-memory processing to parallelize those methods, and compared our results with the popularly used Hadoop MapReduce-based implementation. Our experiments ran on six servers under a medium-sized high-performance cloud computing environment. The results demonstrated that our Spark-based in-memory computing yielded better performance compared with Hadoop MapReduce, with an average 5.4 times speed increase in IDW, and 4.8 times improvement in Kriging. In addition, we evaluated the characteristics of our method in terms of central processing unit, memory usage, and network activities.

High-performance parallel approaches for three-dimensional light detection and ranging point clouds gridding

Abstract. With the rapid advance of remote sensing technology, the amount of three-dimensional point-cloud data has increased extraordinarily, requiring faster processing in the construction of digital elevation models. There have been several attempts to accelerate the computation using parallel methods; however, little attention has been given to investigating different approaches for selecting the most suited parallel programming model for a given computing environment. We present our findings and insights identified by implementing three popular high-performance parallel approaches (message passing interface, MapReduce, and GPGPU) on time demanding but accurate kriging interpolation. The performances of the approaches are compared by varying the size of the grid and input data. In our empirical experiment, we demonstrate the significant acceleration by all three approaches compared to a C-implemented sequential-processing method. In addition, we also discuss the pros and cons of each method in terms of usability, complexity infrastructure, and platform limitation to give readers a better understanding of utilizing those parallel approaches for gridding purposes.

Optimizing ground return detection through forest canopies with small footprint airborne mapping LiDAR

The capability of airborne LiDAR scanners (ALS) to record returns from the ground surface and other targets occluded by forest canopies has been of great value for geosciences and military operations. In this paper we present preliminary results from efforts aimed to characterize different types of forest canopies and to assess the quantity and quality of potential ground returns obtained through different configurations of small footprint airborne mapping LiDAR systems. The final goal of this work is to provide a methodology that allows for the quantification of the “openness” of a forest canopy and procedures to determine the best configuration of ALS systems that ensures maximum detection of ground returns independent of the many different system designs currently available.

Airborne lidar point cloud-based below-canopy line-of-sight visibility estimator

Abstract Point cloud data collected by small-footprint lidar scanning systems have proven effective in modeling the forest canopy for extraction of tree parameters. Although line-of-sight visibility (LOSV) in complex forests may be important for military planning and search-and-rescue operations, the ability to estimate LOSV from lidar scanners is not well developed. A new estimator of below-canopy LOSV (BC-LOSV) by addressing the problem of estimation of lidar under-sampling of the forest understory is created. Airborne and terrestrial lidar scanning data were acquired for two forested sites in order to test a probabilistic model for BC-LOSV estimation solely from airborne lidar data. Individual crowns were segmented, and allometric projections of the probability model into the lower canopy and stem regions allowed the estimation of the likelihood of the presence of vision-blocking elements for any given LOSV vector. Using terrestrial lidar scans as ground truth, we found an approximate average absolute difference of 20% between BC-LOSV estimates from the airborne and terrestrial point clouds, with minimal bias for either over- or underestimates. The model shows the usefulness of a data-driven approach to BC-LOSV estimation that depends only on small-footprint airborne lidar point cloud and physical knowledge of tree phenology.

A Trail Detection Using Convolutional Neural Network

Small-footprint airborne LiDAR scanning systems are effective in modelling forest structures and can also improve trail detection. We propose a trail detection method through a machine learning method from the LiDAR points. To do that, we analyze features for detecting a trail, digitize each feature and combine the results to distinguish between trail and non-trail areas. Our proposed method shows the feasibility of trail detection by using airborne LiDAR points gathered in dense mixed forest.

Vehicle detection from airborne LiDAR point clouds based on a decision tree algorithm with horizontal and vertical features

ABSTRACT The object-based point cloud analysis (OBPCA) method has been used for vehicle detection from airborne light detection and ranging (LiDAR) point clouds with a relatively simple process and exhibits a degree of accuracy as high as that of a three-dimensional point cloud-based detection scheme. However, it only utilizes horizontal features of the segmented point clouds, and it uses thresholds established by heuristic observation and experience. In this article, we present a novel method for vehicle detection from airborne LiDAR point clouds based on a decision tree algorithm with horizontal and vertical features. It calculates the horizontal and vertical features for segments created by the filtering and segmentation processes, and it establishes a vehicle detection model by training a decision tree classifier with horizontal and vertical features of the segments. Our experiment shows that our proposed method outperforms the previous method in terms of recall and precision by 13.14% and 30.02%, respectively.

A trail detection method using statistical analysis of trail features in dense forest

Small-footprint airborne LiDAR scanning systems are effective in modelling forest structures and can also improve trail detection. We propose a trail detection method through a statistical analysis from the LiDAR points. To do that, we statistically analyze features of trails for detecting a trail and digitized each feature and combine the results to distinguish between trail and non-trail areas. Our proposed method shows the feasibility of trail detection by using airborne LiDAR points gathered in dense mixed forest.

Vehicle Detection Method Based on Object-Based Point Cloud Analysis Using Vertical Elevation Data

Among various vehicle extraction techniques, OBPCA (Object-Based Point Cloud Analysis) calculates features quickly by coarse-grained rectangles from top-view of the vehicle candidates. However, it uses only a top-view rectangle to detect a vehicle. Thus, it is hard to extract rectangular objects with similar size. For this reason, accuracy issue has raised on the OBPCA method which influences on DEM generation and traffic monitoring tasks. In this paper, we propose a novel method which uses the most distinguishing vertical elevations to calculate additional features. Our proposed method uses same features with top-view, determines new thresholds, and decides whether the candidate is vehicle or not. We compared the accuracy and execution time between original OBPCA and the proposed one. The experiment result shows that our method produces 6.61% increase of precision and 13.96% decrease of false positive rate despite with marginal increase of execution time. We can see that the proposed method can reduce misclassification.

Estimation of 2-D Clutter Maps in Complex Under-Canopy Environments From Airborne Discrete-Return Lidar

Detection of near-ground objects occluded by above-ground vegetation from airborne light detection and ranging (lidar) measurements remains challenging. Our hypothesis is that the probability of obstruction due to objects above ground at any location in the forest environment can be reasonably characterized solely from airborne lidar data. The essence of our approach is to develop a data-driven learning scheme that creates high-resolution two-dimensional (2-D) probability maps for obstruction in the under-canopy environment. These maps contain information about the probabilities of obstruction (clutter map) and lidar undersampling (uncertainty map) in the near-ground space. Airborne and terrestrial lidar data and field survey data collected within the forested mountainous environment of Shenandoah National Park, Virginia, USA are utilized to test and evaluate the proposed approach in this work. A newly developed individual tree detection algorithm is implemented to estimate the undersampled stem contributions to the probability of obstruction. Results show the effectiveness of the tree detection algorithm with an accuracy index (AI) of between 61.5% and 80.7% (tested using field surveys). The estimated clutter maps are compared to the maps created from terrestrial scans (i.e., ground truth) and the results show the root-mean-square error (RMSE) of 0.28, 0.32, and 0.34 at three study sites. The overall framework in deriving near-ground clutter and uncertainty maps from airborne lidar data would be useful information for the prediction of line-of-sight visibility, mobility, and above-ground forest biomass.