Earticle

Point clouds have ability to express the spatial entities, however, the point clouds redundancy always involves some uncertainties in computer recognition and model construction. Therefore, point clouds thinning is an indispensable step in point clouds model reconstruction and other applications. To overcome the shortcomings of complex classification index and long time consuming in existing point clouds thinning algorithms, this paper proposes a point clouds fast thinning algorithm. Specifically, the two-dimensional index is established in plane linear array (x, y) for the scanned point clouds, and the thresholds of adjacent point distance difference and height difference are employed to further delete or retain the selected sample point. Sequentially, the index of sample point is traversed forwardly and backwardly until the process of point clouds thinning is completed. The results suggest that the proposed new algorithm can be applied to different targets when the thresholds are built in advance. Besides, the new method also performs superiority in time consuming, modelling accuracy and feature retention by comparing with octree thinning algorithm.

In order to solve the problem of point clouds coordinate conversion of non-directional scanners, this paper proposes a basic Rodrigues rotation method. Specifically, we convert the 6 degree-of-freedom (6-DOF) rotation and translation matrix into the uniaxial rotation matrix, and establish the equation of objective vector conversion based on the basic Rodrigues rotation scheme. We demonstrate the applicability of the new method by using a bar-shaped emboss point clouds as experimental input, the three-axis error and three-term error as validate indicators. The results suggest that the new method does not need linearization and is suitable for optional rotation angle. Meanwhile, the new method achieves the seamless splicing of point clouds. Furthermore, the coordinate conversion scheme proposed in this paper performs superiority by comparing with the iterative closest point (ICP) conversion method. Therefore, the basic Rodrigues rotation method is not only regarded as a suitable tool to achieve the conversion of point clouds, but also provides certain reference and guidance for similar projects.

포인트 클라우드를 활용한 폴리곤 모델 활용에 대해서는 지속적으로 연구되고있다. 이러한 연구는 대부분 하이 폴리곤 모델을 대상으로 생성된 모델을 활용한 기술이다. 오늘날 비디오 게임 등에서 활용되는 하이 폴리곤 모델은 대부분 높은 퍼포먼스를 바탕으로 화려한 효과를 표현한다. 그러나 게임 등의 콘텐츠에 있어서 화려한 효과와 높은 품질의 리소스를 활용한 게임만이 만족도를 충족시켜주는 것이 아니다. 콘텐츠 에 따라서 의도적으로 로우 폴리곤 모델을 활용하기도 한다. 인디 게임 등에서는 낮 은 접근성을 장점으로 로우 폴리곤 형태의 모델을 제작하여 이를 활용한 게임들도 지속적으로 개발되어 왔다. 따라서 본 연구에서는 이러한 변화에 맞춰서 포인트 클라우드 데이터를 활용하여 로우 폴리곤 모델을 생성 및 음영 처리하는 방법을 제안한 다. 여러 개의 모델을 생성하여, 각각의 모델이 하나의 텍스처를 공유하고, 음영을 처리한다. 마지막으로 라이트맵 기법의 결과물과 비교하는 실험을 진행한다.

The use of polygon models using point cloud is continuously studied. These studies are mostly techniques using models generated for high-polygon models. Most high-polygon models used in video games today express spectacular effects based on high performance. However, games that utilize colorful effects and high quality resources are not the only ones that satisfy the satisfaction level. Depending on the content, the low-polygon model may be used intentionally. In indie games, low-polygon models have been produced with low accessibility, and games that utilize them have been continuously developed. Therefore, this study suggests ways to utilize point cloud data to create and shade low polygon models in response to these changes. Generating several models, each model shares one texture and processes shading. Finally, conduct an experiment comparing the results of the light map technique.

This study addresses the challenge of severe class imbalance in semantic segmentation of rural watershed point clouds. Due to the dominance of a few classes and the scarcity of minority classes, segmentation performance can degrade significantly. To mitigate this issue, we investigated loss function strategies based on class weighting and focal loss using the Point Transformer V3 architecture. The dataset, collected via LiDAR from Sanyang Reservoir watershed in Icheon, South Korea, was labeled into ten semantic classes and preprocessed using sub-sampling. Four loss function configurations were evaluated: baseline cross entropy, class-weighted cross entropy loss, focal loss, and a combination of focal loss with class weighting. Experimental results show that combining class weighting and focal loss achieves the best performance (mIoU = 0.3222; OA = 0.7464), outperforming the baseline, with clear IoU improvements in several minority classes such as Road, Water, and House, while Bridge and Reservoir remain challenging due to extreme data scarcity. Hyperparameter tuning further indicates optimal focal loss parameters of γ = 3.0 without class weighting (mIoU = 0.3311) and γ = 2.0 with class weighting (mIoU = 0.3222). These findings highlight the effectiveness and limitations of tailored loss re-weighting for addressing class imbalance in rural watershed point cloud segmentation.

하천에 광범위하게 활착되는 식생은 수위 변화 및 흐름 저항에 절대적인 영향을 미칠 뿐만 아니라 하천 시스템 전반에 영향을 미치는 중요 요소이다. 따라서 유입되는 식생의 형태와 규모를 정확하게 파악하는 것이 매우 중요함에도 불구하고 현장에서 이를 파악하기란 쉽지 않은 일이다. 따라서 최근에는 지상 레이저 스캐닝 등을 활용하여 대용량의 식생 정보를 취득하는 연구가 시도되고 있다. 그러나 식생의 복잡한 형상으로 인해 캐노피 영역의 정확한 정보를 획득하기 어려우며, 자연적인 영향에 매우 민감하게 반응한다는 한계가 있다. 본 연구에서는 3차원 지상 레이저 스캐닝을 통해 수집된 고해상도의 포인트 클라우드 데이터를 복셀 형식으로 재구현하여 식생의 물리적 구조를 분석하였다. 먼저 잎이 없는 단순한 형태, 잎이 있는 완전한 형태의 식생 및 패치 규모 식생 조건으로 설정하여 각각의 물리적 구조를 분석하였다. 이를 위해 측정된 데이터의 이상치 제거 및 불필요한 데이터의 필터링을 위해 통계적 이상치 제거 방법을 활용하여 각각 17 %, 26 %, 25 %의 포인트를 제거하였다. 또한 후처리 된 포인트 클라우드로부터 복셀 크기별 식생 형상을 재구현하여 실제 식생의 부피와 비교하였으며, 분석 결과, 오차 범위는 각 조건별로 8 %, 25 %, 63 %로 나타났다. 대상 샘플의 규모가 클수록 더 큰 오차가 발생하였으며, 복셀 크기 조정 시 식생의 표면이 시각적으로 비슷하게 보이지만 전체 식생의 부피는 이러한 변화에 매우 민감한 것으로 나타났다.

Vegetation affects water level change and flow resistance in rivers and impacts waterway ecosystems as a whole. Therefore, it is important to have accurate information about the species, shape, and size of any river vegetation. However, it is not easy to collect full vegetation data on-site, so recent studies have attempted to obtain large amounts of vegetation data using terrestrial laser scanning (TLS). Also, due to the complex shape of vegetation, it is not easy to obtain accurate information about the canopy area, and there are limitations due to a complex range of variables. Therefore, the physical structure of vegetation was analyzed in this study by reconfiguring high-resolution point cloud data collected through 3-dimensional terrestrial laser scanning (3D TLS) in a voxel. Each physical structure was analyzed under three different conditions: a simple vegetation formation without leaves, a complete formation with leaves, and a patch-scale vegetation formation. In the raw data, the outlier and unnecessary data were filtered and removed by Statistical Outlier Removal (SOR), resulting in 17%, 26%, and 25% of data being removed, respectively. Also, vegetation volume by voxel size was reconfigured from post-processed point clouds and compared with vegetation volume; the analysis showed that the margin of error was 8%, 25%, and 63% for each condition, respectively. The larger the size of the target sample, the larger the error. The vegetation surface looked visually similar when resizing the voxel; however, the volume of the entire vegetation was susceptible to error.

Data registration is a common operation in object digitalization. This paper proposes an algorithm for registering point cloud data with such quadratic surfaces as planes and cylinders. In the proposed algorithm, the variation of neighboring points’ normal vectors are utilized to detect data that is located at the intersection of different quadratic surfaces, which is then deleted from the point clouds. Next, region segmentation is done on remaining point clouds through the growth algorithm, and feature descriptors are created for each detected quadratic surface. Finally, rough registration of data is achieved via matching of feature descriptors, and then fine registration is achieved by optimizing rough registration results through the ICP algorithm. Experimental results show that the algorithm is insensitive to outliers, has low requirements on the original poses of data and that its registration success rate is higher than that of the original ICP algorithm.

The registration of point cloud is a key problem for model acquisition and 3D reconstruction, it was Put forward that a kind of automatic registration method from multi-views depth image to complete geometric models. First, According to the invariant characteristics of the relative position of the space points in the condition of rigid body transformation, the effective initial matching points array was construct with curvature invariant features and ZNCC, and the coordinate transformation of the matching feature points was solved based on the Unit quaternion, thus the data coarse registration was completed; Then by using fine matching technology based on adaptive distance function and the improved iterative closest surface, the different perspectives of clouds were optimally matched in 3D space; Finally registration error was calculated according to the matching results, and the registration accuracy and speed were analyzed. The results show that, the method can effectively improve the efficiency of registration in the guarantee of the accuracy of registration.

In order to remove redundant data and resolve conflicts in point clouds, we proposed fuzzy clustering reduction strategy in this paper. Original point clouds are decreased before computing other pretreatments. The proposed method involves three processes: reduction of the original data using fuzzy clustering while the point clouds are divided into sub-domains using octree structure, generation of the sub-surface that is fitted the sub-surface by implicit function in each sub-domain, the normal alignment that are computed normal of sub-surface and inference the global normal of surface using iteratively propagate algorithm. The method is suitable to reduce mass point clouds to reconstruct surface that can keep the property of surface. The experimental results show that the model with less sharp feature is more effective than complex model to reduce point clouds by fuzzy clustering.

With the reverse engineering development, the accuracy of system is more important in reconstruction, especially in non-contacting measurement. This paper provides a new method measuring the accuracy the point clouds, define a image probability and the point probability according to uncertainty data. The quantity of the uncertain point data is important to measuring the result of reconstruction. The prior data can be catch from the last measurement process, especially the edge data or characteristic points. Referring to prior data, basing on the Bayesian theory the more accuracy posterior data can be computed in this paper. We divided the point cloud into different areas, and organized the data with hierarchical tree-structure. According to the probability of one tree node, we adjust the area corresponding to the node. At last, by using the existing experimental equipment, we verify the measurement of point cloud accuracy algorithm. The depth data was obtained by a laser scanner---SICK LMS100. The depth data can be computed as point data with uncertainty. The result of the reconstruction deeply relies on the quality of prior data.

This paper presents a fuzzy normal estimate for mass point clouds of irregular models in reconstruction. The irregular model is complex object that some part is smooth and some parts are irregular including sharp features. Therefore, we put kNN and curvature of mass point clouds to fuzzy inference system to divide the kind of point clouds and the output of FIS can determine which part of tooth point clouds belong to. For different kinds point clouds, corresponding algorithm is given. Point clouds in the smooth area are estimated normal by PCA directly and ones in other regions of thin or sharp area are estimated by checker and attach points. This method is simpler than those complex methods used on the whole point clouds directly. The experiment results show that much time is saved and surface reconstruction is very fine than PCA and WLOP.

According to mass point clouds without explicit topology relation, a parallel fast sort algorithm is proposed in this paper. Morton order is introduced and used to merge one-dimensional data. The mass point clouds of irregular model are generated corresponding address code named Morton code and these points are stored in the octree structure chain. And then a parallel fast sort algorithm based on Euclidean distance is used to sort by CPU and GPU. The k-Nearest Neighbors of point can be located in the chain. The experiment results show that much time is saved and k-Nearest Neighbors of point can be searched directly. This algorithm is simpler than those complex sort methods used on the whole point clouds.

This paper presents fuzzy classification strategy for the hole-filling that can classify the incomplete mass point clouds and improve the precision. The irregular model is complex object that some part is smooth and some parts are irregular including sharp features. Therefore, we put kNN and curvature of mass point clouds to fuzzy inference system to divide the type of the hole of mass point clouds and the output of FIS can determine which part of point clouds belong to. For different kind holes, corresponding algorithm is given. Point clouds in the smooth area are reconstructed by implicit directly and ones in other regions of thin or sharp area are reconstructed by attach points. This method is simpler than those complex methods used on the whole point clouds directly. The experiment results show that classification can save much time and surface reconstruction is very fine.

This paper presents a self-organized fuzzy neural network (SOFNN) surface reconstruction algorithm suitable for point clouds without normal. It overcomes the defect of traditional Delaunay triangulation which is difficult to reconstruct point clouds with noises and implicit function which is limited to the number of point clouds and point clouds are required very strict. The SOFNN is based on the fuzzy clustering method optimizing training data before learning fuzzy rules, in order to remove noise data and resolve conflicts in data. The approach not only reduce computational burden of neural network, but also make it easy to fit the surface for point clouds without normal and suitable for mass point clouds. The feature of the SOFNN has dynamic self-organized structure, fast learning speed and flexibility in learning. The experiment results show that is very fine.

High-performing architecture should be designed by taking into account the mutual dependency between the new building and the local context. The performative architecture plays an important role to avert any unforeseen failures after the building has been built; particularly ones related to the microclimate impacts that affect the human comfort. The use of the concept of solar envelopes helps designers to construct the developable mass of the building design considering the solar access and the site obstruction. However, the current analysis method using solar envelopes lack in terms of integrating the detailed information of the existing context during the simulation process. In architectural design, often the current site modelling not only absent in preserving the complex geometry but also information on the surface characteristics. Currently, the emerging applications of point clouds offer a great possibility to overcome these limitations, since they include the attribute information such as XYZ as the position information and RGB as the color information. This study particularly presents a comparative analysis between the manually built 3D models and the models generated from the point cloud data. The modelling comparisons focus on the relevant factors of solar radiation and a set of simulation to calculate the performance indicators regarding selected portions of the models. The experimental results emphasize an introduction of the design approach and the dataset visibility of the 3D existing environments. This paper ultimately aims at improving the current architectural decision of support environment means, by increasing the correspondence between the digital models for performance analysis and the real environments (context of design) during the conceptual design phase.

Coastal erosion poses a serious threat to ecosystems, infrastructure, and coastal livelihoods. Traditional monitoring using satellite or aerial imagery often struggles to distinguish between forest, beach, and sea regions, particularly in complex coastal environments. This study aims to enhance shoreline monitoring accuracy by introducing a LiDAR-based multiclass classification framework that effectively differentiates these terrain types. The proposed method integrates geometric and color-based features extracted from UAV-based LIDAR data collected in Khlung, Chanthaburi, Thailand. Six features were derived from segmented point cloud grids: elevation variability (v1), directional slope variability (v2), and the mean RGB values (Rmean, Gmean, Bmean, RGBmean). Four machine learning models, Random Forest, Decision Tree, Support Vector Machine, and Logistic Regression, were evaluated using cross-validation on 1500 samples. Results indicate that the Random Forest classifier achieved the highest accuracy of 99.78%, outperforming other models. The proposed approach effectively separates forest, beach, and sea regions, reducing misclassification errors and improving the reliability of coastal terrain classification. This study demonstrates the potential of combining LiDAR point cloud geometry and radiometric features for robust multiclass classification. The findings support more precise and automated coastal erosion analysis, offering a foundation for future integration with deep learning and hybrid image point cloud systems.

In this study, we suggest a new segmentation algorithm for processing airborne laser point cloud data which is more memory efficient and faster than previous approaches. The main principle is the reading of data points along a scan line and their direct classification into homogeneous groups as a single process. The results of our experiments demonstrate that the algorithm runs faster and is more memory efficient than previous approaches. Moreover, the segmentation accuracy is generally acceptable.

Recently, quantitative and repetitive inspections of the old urban area were conducted because many structures exceed their designed lifetime. The health of a building can be validated from the condition of the outer wall, while the slant angle of the wall widely serves as an indicator of urban regeneration projects. Mostly, the inspector directly measures the inclination of the wall or partially uses 3D point measurements using a static light detection and ranging (LiDAR). These approaches are costly, time-consuming, and only limited space can be measured. Therefore, we propose a mobile mapping system and automatic slant map generation algorithm, configured to capture urban environments online. Additionally, we use the LiDAR-inertial mapping algorithm to construct raw point clouds with gravity information. The proposed method extracts walls from raw point clouds and measures the slant angle of walls accurately. The generated slant angle map is evaluated in indoor and outdoor environments, and the accuracy is compared with real tiltmeter measurements.

In this paper, we propose a novel method for road recognition using 3D point clouds based on a Markov random field (MRF) framework in unstructured and complex road environments. The proposed method is focused on finding a solution for an analysis of traversable regions in challenging environments without considering an assumption that has been applied in many past studies; that is, that the surface of a road is ideally flat. The main contributions of this research are as follows: (a) guidelines for the best selection of the gradient value, the average height, the normal vectors, and the intensity value and (b) how to mathematically transform a road recognition problem into a classification problem that is based on MRF modeling in spatial and visual contexts. In our experiments, we used numerous scans acquired by an HDL-64E sensor mounted on an experimental vehicle. The results show that the proposed method is more robust and reliable than a conventional approach based on a quantity evaluation with ground truth data for a variety of challenging environments.

This paper presents a practical approach for surface reconstruction from unoriented point clouds. Instead of estimating local surface orientation, we first generate a set of depth images from the input point clouds, and a coarse mesh is then generated based on them by space carving techniques. The resultant mesh is progressively refined by local mesh refinement and optimization according to surface distance measure. A manifold mesh approximating the input points within an given tolerance is finally obtained. Our approach is easy to implement, but has the ability to outputs high quality meshes in different resolutions. We show that the proposed approach is not sensitive to several types of data disfigurement and is able to reconstruct models robustly from variance input data.

3D point clouds are a crucial data format for accurately capturing geometric information in large-scale industrial environments such as shipyards. Deep learning-based object detection technology using 3D point clouds enables automated production management and process optimization. However, the large volume characteristic of 3D point clouds remains a challenge due to the resources and time required for data processing and dataset construction. The large volume of 3D point clouds leads to excessive computational costs, storage demands, and time consumption during dataset construction and training. Therefore, it is necessary to appropriately reduce the dataset size for efficient utilization while ensuring object detection performance. This necessitates a study on dataset downsampling strategies that maintain optimal density and detection accuracy. In this study, an experimental dataset similar to the S3DIS (Stanford Large-Scale 3D Indoor Spaces) dataset was constructed. The density of the 3D point clouds was adjusted in five levels by reducing points per unit area by 20% increments. These datasets were applied to a deep learning architecture to analyze object detection accuracy. Subsequently, the findings were applied to a shipyard dataset to streamline large volume point clouds and evaluate detection performance, thereby assessing their practical applicability. The results demonstrated that reducing the experimental dataset density to approximately 20% still maintained object detection accuracy of around 95% IoU for key objects. This indicates that lightweight datasets can reduce processing resources and costs while preserving detection performance. Additionally, applying the approach to real shipyard datasets revealed that object detection was feasible with reduced data (approximately 4.6% of the raw data). This study provides a practical framework for constructing efficient deep learning models for object detection by downsampling datasets in large-scale industrial environments like shipyards. It is expected to contribute to the establishment of automated data management systems for production management and process efficiency enhancement. Further analysis is required to evaluate performance at extreme low densities (below 20%). Moreover, while this study employed simple downsampling techniques, future work should explore the performance of various downsampling methods to optimize detection accuracy.