Terrestrial laser scanning (TLS) is an important technique for tree stem detection. In this paper, a point-based method for stem detection is proposed using single-scan TLS data. One of the main concerns is the point density, which decreases rapidly with the increasing distance to the scanner position. In the proposed method, the search radius is generated adaptively, based on the relationship between the distance and point density, to make sure that the neighborhood maintains a similar scale to the corresponding point density. The belonging of each point is recognized with cuckoo search-based support vector machine, and the points labeled as stem are then clustered and filtered for further verification. The threshold for the small cluster filtering is also adaptive to deal with the problem of the cluster point number decreasing as a function of distance. The stem position is calculated with the lowest cylinder from the cluster segmentation and modeling for the stem mapping. Experiments were carried out on two plots with radii of more than 130 m. The overall detection rate was 76.1%, and 75% of the stems outside 80 m were detected with the adaptive radius, despite the point density being less than 5 cm.

This paper proposes a novel method to reconstruct hierarchical 3D tree models from Mobile Laser Scanning (MLS) point clouds. Starting with a neighborhood graph from the tree point clouds, the method treats the root point of the tree as a source point and determines an initial tree skeleton by using the Dijkstra algorithm. The initial skeleton lines are then optimized by adjusting line connectivity and branch nodes based on morphological characteristics of the tree. Finally, combined with the tree point clouds, the radius of each branch skeleton node is estimated and flat cones are used to simulate tree branches. A local triangulation method is used to connect the gaps between two joint flat cones. Demonstrated by street trees of different sizes and point densities, the proposed method can extract street tree skeletons effectively, generate tree models with higher fidelity, and reconstruct trees with different details according to the skeleton level. It is found out the tree modeling error is related to the average point spacing, with a maximum error at the coarsest level 6 being about 0.61 times the average point spacing. The main source of the modeling error is the self-occlusion of trees branches. Such findings are both theoretically and practically useful for generating high-precision tree models from point clouds. The developed method can be an alternative to the current ones that struggle to balance modeling efficiency and modeling accuracy.

In indoor scenes, pose estimation drift in SLAM systems is challenging to mitigate due to factors like low texture and texture repetition. Several studies utilize Manhattan structural information to achieve low-drift rotation estimation consequently reducing the cumulative error. However, the application scenarios of these methods are limited by the Manhattan assumption. To address this problem, we propose a robust RGBD odometry for tracking in more general structured scenes, which can represent a wider range of scenes by leveraging Atlanta World assumption, consisting of a vertical direction and multiple horizontal directions. To this end, we design tracking by detection algorithms for extracting Key Local Manhattan World in each frame, which is defined as the most stable structural information in current frame. Specifically, we detect local Manhattan Worlds in current frame, if a more stable structural feature emerges, it becomes the new tracking target. The angle between these Manhattan Worlds are then computed, constructing the Atlanta World. For accurate pose estimation, we use Key Local Manhattan World to first estimate low-drift rotation followed by estimating translation using point-line structural features. Extensive experiments on public benchmark and self-recorded datasets show that our method outperforms existing state-of-the-art methods with a significant margin of 22% while extending their applicability to the Atlanta World.

Point cloud classification is a crucial procedure in ground scene interpretation. In this letter, density-adaptive geometric features are proposed for the classification of terrestrial laser scanning data, with the problem brought by point density variation as one of the main concerns. For each point, point spacing is estimated, respectively, based upon the distance to scanner position and the angular resolution, and then used as neighborhood scale basis to generate the search range of optimal radius. In feature extraction, we modify some common geometric features to adapt to density variation, e.g., a polar projection grid is proposed to generate projection features instead of commonly used rectangular grid. The polar grid can make sure similar number of laser beams passing through each grid. An evaluation involving five classifiers is carried out in an outdoor scene captured by Reigl-VZ400 scanner and the results show density-adaptive features have better and more stable performances than features without considering density variation, with the highest overall accuracy of 95.95%. Moreover, the proposed features perform well on the recognition of buildings from a large distance (more than 300 m in this letter).

Accurate road information is important for applications involving road maintenance, intelligent transportation, and road network updates. Mobile laser scanning (MLS) can effectively extract road information. However, accurately extracting road edges based on large-scale data for complex road conditions, including both structural and non-structural road types, remains difficult. In this study, a robust method to automatically extract structural and non-structural road edges based on a topological network of laser points between adjacent scan lines and auxiliary surfaces is proposed. The extraction of road and curb points was achieved mainly from the roughness of the extracted surface, without considering traditional thresholds (e.g., height jump, slope, and density). Five large-scale road datasets, containing different types of road curbs and complex road scenes, were used to evaluate the practicality, stability, and validity of the proposed method via qualitative and quantitative analyses. Measured values of the correctness, completeness, and quality of extracted road edges were over 95.5%, 91.7%, and 90.9%, respectively. These results confirm that the proposed method can extract road edges from large-scale MLS datasets without the need for auxiliary information on intensity, image, or geographic data. The proposed method is effective regardless of whether the road width is fixed, the road is regular, and the existence of pedestrians and vehicles. Most importantly, the proposed method provides a valuable solution for road edge extraction that is useful for road authorities when developing intelligent transportation systems, such as those required by self-driving vehicles.