In environmental protection, tree monitoring plays an essential role in maintaining and improving ecosystem health. However, precise monitoring is challenging because existing datasets fail to capture continuous fine-grained changes in trees due to low-resolution images and high acquisition costs. In this paper, we introduce UAVTC, a large-scale, long-term, high-resolution dataset collected using UAVs equipped with cameras, specifically designed to detect individual Tree Changes (TCs). UAVTC includes rich annotations and statistics based on biological knowledge, offering a fine-grained view for tree monitoring. To address environmental influences and effectively model the hierarchical diversity of physiological TCs, we propose a novel Hyperbolic Siamese Network (HSN) for TC detection, enabling compact and hierarchical representations of dynamic tree changes. Extensive experiments show that HSN can effectively capture complex hierarchical changes and provide a robust solution for fine-grained TC detection. In addition, HSN generalizes well to cross-domain face anti-spoofing task, highlighting its broader significance in AI. We believe our work, combining ecological insights and interdisciplinary expertise, will benefit the community by offering a new benchmark and innovative AI technologies. Source code is available on https://github.com/liyantett/Tree-Changes-Detection-with-Siamese-Hyperbolic-network.

Aerial and terrestrial laser scanning (TLS) technologies offer complementary, high-precision 3-D data on forest structure. Registering point cloud data from multiplatform is crucial for achieving a more comprehensive understanding of forest structure. Currently, multisource forest point cloud registration remains challenging due to factors such as unstable point- and object-level features, varying observation perspectives, and nonstandardized processing pipelines. To address these challenges, this study introduces a unified and automated method for registering aerial and ground-based point clouds in forest areas. First, keypoints are extracted from multisource point clouds using fuzzy c-means (FCM) clustering. Local transformation is then derived from the keypoint sets using the gradient-based local convergence (GLC) algorithm, which is integrated with a nested branch and bound (BnB) structure for further optimization. We also developed a novel distribution similarity index (DSI) to evaluate the alignment of the keypoint sets and determine the initial transformation. Finally, this initial transformation is applied to the ground-based point cloud and refined using the GLC algorithm. Compared with existing methods that rely on point and object features, the proposed method does not depend on geometric descriptions or tree attributes (e.g., tree position and canopy structure). It also demonstrates robustness to variations in the initial position of the point clouds. Testing on 15 datasets with varying plot sizes and tree characteristics shows that the proposed method achieves comparable or superior performance to existing methods, with an average distance residual of 6.59 cm and an average runtime of 90.04 s.

Precise positioning is fundamental to the internet of things that delivers insights into everything from large-scale business to ordinary smart life. Accurate localization and positioning in global navigation satellite system (GNSS) denied environments, such as indoor-, underground-spaces, and forests, is one of the most prosperous research fields because of the great complexity prompted by various challenging application scenarios. Different sensors, algorithms, and combinations of those have been developed in past decades, which provided a great variety of possible solutions that deliver different positioning accuracies. However, a rigorous evaluation of the positioning accuracy of different mainstream solutions is missing, mainly because of the difficulties in acquiring reliable ground truth for referencing and the lack of comparable test/application conditions. A comprehensive benchmarking was carried out in this study based on the comparisons of six solutions that consist of different combinations of five positioning technologies, i.e., 1) ultra-wideband (UWB) and inertial measurement unit (IMU); 2) UWB, IMU, and camera; 3) UWB and light detection and ranging (LIDAR); 4) UWB and radio detection and ranging (RADAR); 5) IMU, camera and LIDAR; and 6) UWB, IMU, camera and LIDAR. The five technologies, i.e., UWB, IMU, camera, RADAR, and LIDAR, were commonly regarded as those that are with high applicability, accuracy, and robustness. New anchors self-positioning algorithm and integrity monitoring algorithm were proposed to further aid the compared solutions and the benchmark. High-precision survey (millimeter) -level ground truth references were acquired at indoor and outdoor test locations and applied in the evaluations, to assist reliable quantitive benchmarks about the positioning accuracies and stabilities of the compared solutions. The strengths, limitations, and potentials of each solution were analyzed. It was revealed that all relative positioning solutions accumulate positioning errors over time. Such accumulation was of the highest significance for RADAR, followed by camera. LIDAR is presented to be the most robust solution for relative positioning. Compared to camera, LIDAR, and RADAR alone, the integration of different technologies clearly improved the performance. The tight-coupling performed slightly superior to loose-coupling, and the unscented Kalman filter with tight-coupling had a higher positioning accuracy in most cases.

Terrestrial laser scanning (TLS) technology has been demonstrated to be able to measure forest structure fast, nondestructively, and with high precision. The registration of point cloud data from different scans is a prerequisite for an in-depth understanding of forest structure. Currently, automated registration methods for forest point clouds typically rely on tree attributes (such as tree position and stem diameter). However, these methods are often suitable for easy forest conditions, where the occlusion effects are less significant and the overlapped areas between scans are large. This article proposes an automated point cloud registration method utilizing ground points without the need for extracting individual tree attributes. In the state-of-the-art hardware setup, the scanner automatically maintains a level position during data acquisition. Therefore, the differences in z-coordinates between the corresponding points in different scans should be equal. Leveraging this property and the rhombic region correspondence principle (RRCP), the proposed method identifies the overlapping rhombic region in the target and reference TLS data, which is located in the middle of the two scanning positions in the plot. Points within the overlapping rhombic region are used as registration primitives. Experiments were carried out in 24 plots with diverse stem densities, tree species, and altitudes located in two test areas in Jiangxi and Zhejiang, China. The resulting average pointwise error, average translation error, and average rotation error of TLS-TLS forest point cloud registrations are 5.77 cm, 4.92 cm, and 3.37 mrad. The results showed the potential of applying ground points for scan-to-scan registration.

First-order branches, which are directly attached to the stem, form the primary structure of an individual tree. They play a crucial role in determining the tree’s photosynthetic efficiency and reproduction capabilities. Until recently, to nondestructively retrieve first-order branches remains a great challenge. Terrestrial laser scanning (TLS) has gained prominence in forest inventories, achieving promising estimations of fundamental metrics. These successes indicate that TLS could be a promising tool to acquire individual tree branch information. Previous studies on branch modeling have predominantly focused on the tree-level or simulation data, with limited occlusion effects. This study addresses the gap between the practical need and current research by developing an automated algorithm to retrieve the first-order branches from the plot-level TLS data and demonstrating the applicability and potential of TLS data in silviculture and ecology applications. The method was tested on 416 trees across six plots, categorized into three complexity levels. The experiment results indicated a varying accuracy of the branch detection and length estimation across different plot complexity levels. Specifically, plot 1 exhibited the highest branch detection accuracy, i.e., 71.42% recall and 73.15% precision, and the highest accuracy of branch length estimation was observed in plots 2 and 3, i.e., approximately 52.90 cm in root-mean-square error (RMSE) and 63.96% in RMSE%. The accuracy of branch detection generally decreased with increased plot complexity, as the average${F}1$-scores of “easy,” “medium,” and “difficult” plots are 64.73%, 63.29%, and 48.63%, respectively. Notably, the proposed method works for the first-order branch modeling at both plot- and individual-tree-levels, though the experiment is primarily based on the plot-level data. This study demonstrates the capability of close-range point clouds in the retrieval of branch structures and underscores its potential in comprehensive forest structure analyses.

Simultaneous Localization and Mapping (SLAM) based on Laser Scanning (LS) has been quickly developed in the last decades. However, the application of LS-SLAM in forest environments is still at an early development stage, limited by the challenges posed by forest environments, such as geometric degeneration and Global Navigation Satellite System (GNSS) denied. The applicability, strengths, and weaknesses of the state-of-the-art LS-SLAM methods has not been investigated and even not been assessed. This study quantitatively evaluated nine state-of-the-art LS-SLAM methods in twelve subtropical forest plots with different levels of complexity, i.e., “Easy”, “Medium”, and “Difficult”. In addition, a robust 3D LS-SLAM method specially designed for real-time forest mapping is proposed. This solution extracts angle-based features and applies a continuous filter in 2D angle image space to identify stable features for enhancing the data alignment performance. The benchmarking results indicated that 1) the LiDAR-only SLAM methods presented an average trajectory accuracy at the 15 cm and 25 cm levels in easy and medium plots, respectively, and failed in difficult plots. 2) the LiDAR-IMU SLAM presented an equivalent or better accuracy in comparison with the LiDAR-only methods, yet their performance was still limited in difficult plots. 3) the SLAM methods with back-end optimization significantly improved the localization results, i.e., with the 10 cm level accuracy in the easy and medium plots, and succeeded in all difficult plots. 4) the trajectory accuracy of the proposed method was at the 5 cm level in all complexity categories. Compared to the state-of-the-art SLAM methods, the trajectory estimation accuracy has doubled, with the RMSE decreasing from the 10 cm to 5 cm level. 5) the inertial measurement unit (IMU) data, properly designed feature extraction method, loop closure, and back-end optimization modules play a crucial role in resolving aggressive motion and enhancing the alignment accuracy and robustness. These outcomes provide valuable guidelines for researchers and practitioners to select existing SLAM solutions in different forest conditions and applications, while, opening a discussion on the future developments that are still necessary.

Efficiently collecting 3D information in complex forest environments holds significant importance for forest surveys. Terrestrial laser scanning or ground-based mobile laser scanning can acquire 3D forest information in relatively smooth terrain conditions, but they still face the challenge of time-consuming and labor consumption. In comparison, small unmanned aerial vehicle (UAV) measurement systems with efficient autonomous exploration capabilities are becoming a new close-range sensing platform for acquiring forest data within intricate canopy environments due to their high-efficiency maneuverability and flexibility. However, achieving efficient autonomous exploration for UAVs remains a significant challenge in complex forest environments. In this paper, we propose a novel autonomous exploration method without relying on the global navigation satellite system. Three heuristic waypoint generation algorithms are introduced to facilitate efficient autonomous exploration of intricate and unknown forest environments using quadrotor. Subsequently, employing nonlinear optimization, B-spline curves are employed for generating smooth, collision-free, and dynamically feasible local planning trajectories. Finally, a sliding window strategy is utilized to quickly adjust the trajectory when obstacles like tree branches are detected. This ensures that the quadrotor can fly without collisions by promptly re-planning its path. We have conducted a series of benchmarked experiments within plantation and natural forest environments. Compared with the classic next-best-view planning (NBVP) method, the proposed method can complete exploration 3-7 times faster than NBVP. Compared with the fast UAV exploration (FUEL) and fast autonomous exploration planner (FAEP) methods, this method reduces the average flight time and distance by 60%. Furthermore, we also validated the effectiveness of the proposed methods in real forest experiments.

For the SLAM system in robotics and autonomous driving, the accuracy of front-end odometry and back-end loop-closure detection determine the whole intelligent system performance. But the LiDAR-SLAM could be disturbed by current scene moving objects, resulting in drift errors and even loop-closure failure. Thus, the ability to detect and segment moving objects is essential for high-precision positioning and building a consistent map. In this paper, we address the problem of moving object segmentation from 3D LiDAR scans to improve the odometry and loop-closure accuracy of SLAM. We propose a novel 3D Sequential Moving-Object-Segmentation (3D-SeqMOS) method that can accurately segment the scene into moving and static objects, such as moving and static cars. Different from the existing projected-image method, we process the raw 3D point cloud and build a 3D convolution neural network for MOS task. In addition, to make full use of the spatio-temporal information of point cloud, we propose a point cloud residual mechanism using the spatial features of current scan and the temporal features of previous residual scans. Besides, we build a complete SLAM framework to verify the effectiveness and accuracy of 3D-SeqMOS. Experiments on SemanticKITTI dataset show that our proposed 3D-SeqMOS method can effectively detect moving objects and improve the accuracy of LiDAR odometry and loop-closure detection. The test results show our 3D-SeqMOS outperforms the existing state-of-the-art methods. We extend the proposed method to the SemanticKITTI: Moving Object Segmentation competition and achieve the 3rd in the leaderboard, showing its effectiveness.

Abstract Interest point detection plays an important role in many computer vision applications. This work is motivated by the light detection and ranging odometry task in autonomous driving. Existing methods are not capable of detecting enough interest points in unstructured scenarios where there are little constructions or trees around, and correspondingly light detection and ranging odometry will fail to continuous localisation. An interest point detector is proposed for detecting interest points from multi‐beam light detection and ranging point cloud using unsupervised convolutional neural network. The point cloud is projected into a two‐dimensional structured data according to the scanning geometry. Then the convolutional neural network filters trained in an unsupervised manner are used to generate a local feature map with the two‐dimensional structured data as input. Finally, interest points are obtained by extracting the grids that have significant differences with their neighbour grids. Based on an odometry benchmark, the experiments show that the proposed interest point detector can capture more local details, which contributes to more than 16% error decrease in point cloud registration in highway scenes.

Fusions of LiDARs (light detection and ranging) and cameras have been effectively and widely employed in the communities of autonomous vehicles, virtual reality and mobile mapping systems (MMS) for different purposes, such as localization, high definition map or simultaneous location and mapping. However, the extrinsic calibration between a camera and a 3D LiDAR is a fundamental prerequisite to guarantee its performance. Some previous methods are inaccurate, have calibration error that is several times the beam divergence, and often require special calibration objects, thereby limiting their ubiquitous use for calibration. To overcome these shortcomings, we propose a novel and high-accuracy method for the extrinsic calibration between a camera and a 3D LiDAR. Our approach relies on the infrared images from a camera with an infrared filter, and the 2D-3D corresponding points in a scene with the corners of a wall can be extracted to calculate the six extrinsic parameters. Experiments using the Velodyne VLP-16 sensor show that the method can achieve an extrinsic accuracy at the level of the beam divergence, which is fully analyzed and validated from two different aspects. Therefore, the calibration method in this paper is highly accurate, effective and does not require special complicated calibration objects; thus, it meets the requirements of practical applications.

The knowledge on the species of individual trees is ineluctable for accurate forest parameter estimation and related studies. Terrestrial Laser Scanning (TLS) remote sensing systems acquire a huge number of point samples that contain very accurate and detailed three dimensional (3D) information of tree structures. Every tree species has unique internal and external crown structural characteristics that can be modeled from its TLS data. However, methods in the state of the art show reduced performance due to inaccurate modeling of tree structures such as the crown, and the branch, and poor selection of features. The proposed method leverages on the fine internal and external crown structural information in TLS data to achieve species classification. We remove noise and stem points in TLS data using a novel voxel neighborhood density-based technique. Internal and external crown geometric features derived from the branch level, and the crown level, respectively, are provided to a non linear Support Vector Machines (SVM) to achieve species classification, and evaluate feature relevance. All experiments were conducted on a set of 75 manually delineated trees belonging to the Spruce, the Pine, and the Birch species.

One of the biggest challenges in forestry research is the effective and accurate measuring and monitoring of forest variables, as the exploitation potential of forest inventory products largely depends on the accuracy of estimates and on the cost of data collection. This paper presented a novel computational method of low-cost forest inventory using global navigation satellite system (GNSS) signals in a crowdsourcing approach. Statistical features of GNSS signals were extracted from widely available GNSS devices and were used for predicting forest attributes, including tree height, diameter at breast height, basal area, stem volume, and above-ground biomass, in boreal forest conditions. The basic evidence of the predictions is the physical correlations between forest variables and the responses of GNSS signals penetrating through the forest. The random forest algorithm was applied to the predictions. GNSS-derived prediction accuracies were comparable with those of the most accurate 2-D remote sensing techniques, and the predictions can be improved further by integration with other publicly available data sources without additional cost. This type of crowdsourcing technique enables the collection of up-to-date forest data at low cost, and it significantly contributes to the development of new reference data collection techniques for forest inventory. Currently, field reference can account for half of the total costs of forest inventory.

Canopy structure plays an essential role in biophysical activities in forest environments. However, quantitative descriptions of a 3-D canopy structure are extremely difficult because of the complexity and heterogeneity of forest systems. Airborne laser scanning (ALS) provides an opportunity to automatically measure a 3-D canopy structure in large areas. Compared with other point cloud technologies such as the image-based Structure from Motion, the power of ALS lies in its ability to penetrate canopies and depict subordinate trees. However, such capabilities have been poorly explored so far. In this paper, the potential of ALS-based approaches in depicting a 3-D canopy structure is explored in detail through an international benchmarking of five recently developed ALS-based individual tree detection (ITD) methods. For the first time, the results of the ITD methods are evaluated for each of four crown classes, i.e., dominant, codominant, intermediate, and suppressed trees, which provides insight toward understanding the current status of depicting a 3-D canopy structure using ITD methods, particularly with respect to their performances, potential, and challenges. This benchmarking study revealed that the canopy structure plays a considerable role in the detection accuracy of ITD methods, and its influence is even greater than that of the tree species as well as the species composition in a stand. The study also reveals the importance of utilizing the point cloud data for the detection of intermediate and suppressed trees. Different from what has been reported in previous studies, point density was found to be a highly influential factor in the performance of the methods that use point cloud data. Greater efforts should be invested in the point-based or hybrid ITD approaches to model the 3-D canopy structure and to further explore the potential of high-density and multiwavelengths ALS data.

Stereo images have long been the main practical data source for the high-accuracy retrieval of 3-D information over large areas. However, stereoscopy has been surpassed by laser scanning (LS) techniques in recent years, particularly in forested areas, because the reflection of laser points from object surfaces directly provides 3-D geometric features and because the laser beam has good penetration capacity through forest canopies. In the last few years, image-based point clouds have become a more widely available data source because of advances in matching algorithms and computer hardware. This paper explores the possibility of using consumer cameras for forest field data collection and presents an application of terrestrial image-based point clouds derived from a handheld camera to forest plot inventories. In the experiment, the sample forest plot was photographed in a stop-and-go mode using different routes and camera settings. Five data sets were generated from photographs taken in the field, representing different photographic conditions. The stem detection accuracy ranged between 60% and 84%, and the root-mean-square errors of the estimated diameters at breast height were between 2.98 and 6.79 cm. The performance of image-based point clouds in forest data collection was compared with that of point clouds derived from two LS techniques, i.e., terrestrial LS (the professional level) and personal LS (an emerging technology). The study indicates that the construction of image-based point clouds of forest field data requires only low-cost, low-weight, and easy-to-use equipment and automated data processing. Photographic measurement is easy and relatively fast. The accuracy of tree attribute estimates is close to an acceptable level for forest field inventory but is lower than that achieved with the tested LS techniques.

Terrestrial laser scanning (TLS) has been demonstrated to be an efficient measurement method in plot-level forest inventories. A permanent sample plot in national forest inventories is typically a small area of forest with a radius of approximately 10 m. In practice, whether reference data can be automatically and accurately collected for larger plot sizes is of great interest. It is expensive to collect references in large areas utilizing conventional measurement tools. The application of static TLS is a possible choice but is very challenging due to its lack of mobility. In this letter, a mobile laser scanning (MLS) system was tested and its implications for forest inventories were discussed. The system is composed of a high performance laser scanner, a navigation unit, and a six-wheeled all-terrain vehicle. In this experiment, about 0.4 ha forest area was mapped utilizing the MLS system. The stem mapping accuracy was 87.5%; the root mean square errors of the estimations of the diameter at breast height and the location were 2.36 cm and 0.28 m, respectively. These results indicate that the MLS system has the potential to accurately map large forest plots and further research on mapping accuracy and cost-benefit analyses is needed.

This paper reports on a study of measuring stem curves of standing trees of different species and in different growth stages using terrestrial laser scanning (TLS). Pine and spruce trees are scanned using the multiscan approach in the field, and trees are felled to measure them destructively for the purpose of obtaining reference values. The stem curves are automatically retrieved from laser point clouds, resulting in an accuracy of ~i1 cm. The corresponding manual measurements yield similar accuracy but fewer measurements at the upper parts of tree stems, compared with the automated measurements. The stem volumes based on stem curve data and field measurements and the best Finnish national allometric volume equations (using tree species, height, and diameters at heights of 1.3 and 6 m as predictors) result in similar accuracy. The measurement accuracy of the stem curves and stem volumes is similar for both pine and spruce trees. The results of this paper confirm the feasibility of using TLS to produce stem curve data in an automated, accurate and noninvasive way and indicate that the point cloud provides adequate information to accurately derive stem volumes from standing trees. The stem curves and volumes retrieved from point clouds can be employed in various forest management activities, such as the calibration of national or regional allometric curve functions and the prediction of profits in preharvest inventories.

The demand for detailed ground reference data in quantitative forest inventories is growing rapidly, e.g., to improve the calibration of the developed models of airborne-laser-scanning-based inventories. The application of terrestrial laser scanning (TLS) in the forest has shown great potential for improving the accuracy and efficiency of field data collection. This paper presents a fully automatic stem-mapping algorithm using single-scan TLS data for collecting individual tree information from forest plots. In this method, the stem points are identified by the spatial distribution properties of the laser points, the stem model is built up of a series of cylinders, and the location of the stem is estimated by the model. The experiment was performed on nine plots with 10-m radius. The stem-location maps measured in the field by traditional methods were used as the ground truth. The overall stem-mapping accuracy was 73%. The result shows that, in a relatively dense managed forest, the majority of stems can be located by the automatic algorithm. The proposed method is a general solution for stem locating where particular plot knowledge and data format are not required.