• ETRI Journal
• /
• v.43 no.4
• /
• pp.594-602
• /
• 2021

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.

• ETRI Journal
• /
• v.44 no.6
• /
• pp.1034-1046
• /
• 2022

The learning-based multiview stereo (MVS) methods for three-dimensional (3D) reconstruction generally use 3D volumes for depth inference. The quality of the reconstructed depth maps and the corresponding point clouds is directly influenced by the spatial resolution of the 3D volume. Consequently, these methods produce point clouds with sparse local regions because of the lack of the memory required to encode a high volume of information. Here, we apply the atrous spatial pyramid pooling (ASPP) module in MVS methods to obtain dense feature maps with multiscale, long-range, contextual information using high receptive fields. For a given 3D volume with the same spatial resolution as that in the MVS methods, the dense feature maps from the ASPP module encoded with superior information can produce dense point clouds without a high memory footprint. Furthermore, we propose a 3D loss for training the MVS networks, which improves the predicted depth values by 24.44%. The ASPP module provides state-of-the-art qualitative results by constructing relatively dense point clouds, which improves the DTU MVS dataset benchmarks by 2.25% compared with those achieved in the previous MVS methods.

• International Journal of CAD/CAM
• /
• v.5 no.1
• /
• pp.19-27
• /
• 2005

As three-dimensional range scanners make large point clouds a more common initial representation of real world objects, a need arises for algorithms that can efficiently process point sets. In this paper, we present a method for extracting smooth surfaces from dense point clouds. Given an unorganized set of points in space as input, our algorithm first uses principal component analysis to estimate the surface variation at each point. After defining conditions for determining the geometric compatibility of a point and a surface, we examine the points in order of increasing surface variation to find points whose neighborhoods can be closely approximated by a single surface. These neighborhoods become seed regions for region growing. The region growing step clusters points that are geometrically compatible with the approximating surface and refines the surface as the region grows to obtain the best approximation of the largest number of points. When no more points can be added to a region, the algorithm stores the extracted surface. Our algorithm works quickly with little user interaction and requires a fraction of the memory needed for a standard mesh data structure. To demonstrate its usefulness, we show results on large point clouds acquired from real-world objects.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.39 no.2
• /
• pp.73-79
• /
• 2021

Recently there are growing interests on the energy conservation and emission reduction. In the fields of architecture and civil engineering, the energy monitoring of structures is required to response the energy issues. In perspective of thermal monitoring, thermal images gains popularity for their rich visual information. With the rapid development of the drone platform, aerial thermal images acquired using drone can be used to monitor not only a part of structure, but wider coverage. In addition, the stereo photogrammetric process is expected to generate 3D point cloud with thermal information. However thermal images show very poor in resolution with narrow field of view that limit the use of drone-based thermal photogrammety. In the study, we aimed to generate 3D thermal point cloud using visible and thermal images. The visible images show high spatial resolution being able to generate precise and dense point clouds. Then we extract thermal information from thermal images to assign them onto the point clouds by precisely establishing photogrammetric collinearity between the point clouds and thermal images. From the experiment, we successfully generate dense 3D thermal point cloud showing 3D thermal distribution over the building structure.

• International Journal of CAD/CAM
• /
• v.7 no.1
• /
• pp.103-112
• /
• 2007

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.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.11a
• /
• pp.109-111
• /
• 2022

일반적인 포인트 클라우드(Point Clouds)는 3 차원 공간상의 포인트가 한 개의 색상 정보만을 포함하고 있는 반면에 플렌옵틱 포인트 클라우드(Plenoptic Point Clouds)는 사실감을 향상시키기 위해 한 개의 포인트가 여러 시점에서 촬영된 색상 정보들을 모두 포함하고 있는 새로운 방식의 볼륨 메트릭 데이터 표현 방식이다. 하지만, 일반적인 포인트 클라우드에 비해 더 많은 색상 정보를 필요로 하기 때문에 효율적인 압축이 필수적이다. 따라서, 본 논문에서는 비디오 기반 포인트 클라우드 압축 표준 기술인 V-PCC 를 기반으로 플렌옵틱 포인트 클라우드의 색상 속성간 중복성 제거를 통해 효율적으로 색상 정보를 압축할 수 있는 방법을 제안한다. 실험 결과 제안 방법은 다중 플렌옵틱 색상 속성 정보를 독립적으로 부호화 경우에 비해 상당한 성능 향상이 있음을 보여준다.

• Korean Journal of Remote Sensing
• /
• v.35 no.5_1
• /
• pp.727-736
• /
• 2019

In this paper, we present how to create a point cloud for a target area using multiple unmanned aerial vehicle images and to remove the gaps and overlapping points between datasets. For this purpose, first, IBA (Incremental Bundle Adjustment) technique was applied to correct the position and attitude of UAV platform. We generate a point cloud by using MDR (Multi-Dimensional Relaxation) matching technique. Next, we register point clouds based on observation vectors between stereo images by doing this we remove gaps between point clouds which are generated from different stereo pairs. Finally, we applied an occupancy grids based integration algorithm to remove duplicated points to create an integrated point cloud. The experiments were performed using UAV images, and our experiments show that it is possible to remove gaps and duplicate points between point clouds generated from different stereo pairs.

• Journal of the Korea Computer Graphics Society
• /
• v.24 no.5
• /
• pp.11-19
• /
• 2018

In this paper, we propose an algorithm for improving the accuracy and rate of convergence when two point clouds with noise and a low overlapping area are registered to each other. We make the most use of the geometric information of the underlying geometry of the point clouds with noise for better accuracy. We select a reasonable region from the noisy point cloud for registration and combine a modified acceleration algorithm to improve its speed. The conventional accuracy improvement method was not possible in a lot of noise, this paper resolves the problem by selecting the reasonable region for the registration. And this paper applies acceleration algorithm for a clone to low overlap point cloud pair. A simple algorithm is added to the conventional method, which leads to 3 or 4 times faster speed. In conclusion, this algorithm was developed to improve both the speed and accuracy of point cloud registration in noisy and low overlap case.

• Journal of the Society of Naval Architects of Korea
• /
• v.59 no.5
• /
• pp.330-337
• /
• 2022

Research on object detection algorithms using 2D data has already progressed to the level of commercialization and is being applied to various manufacturing industries. Object detection technology using 2D data has an effective advantage, there are technical limitations to accurate data generation and analysis. Since 2D data is two-axis data without a sense of depth, ambiguity arises when approached from a practical point of view. Advanced countries such as the United States are leading 3D data collection and research using 3D laser scanners. Existing processing and detection algorithms such as ICP and RANSAC show high accuracy, but are used as a processing speed problem in the processing of large-scale point cloud data. In this study, PointNet a representative technique for detecting objects using widely used 3D point cloud data is analyzed and described. And RandLA-Net, which overcomes the limitations of PointNet's performance and object prediction accuracy, is described a review of detection technology using point cloud data was conducted.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.528-529
• /
• 2023

Due to vast points and irregular structure, labeling full points in large-scale point clouds is highly tedious and time-consuming. To resolve this issue, we propose a novel point-based transformer network in weakly-supervised semantic segmentation, which only needs 0.1% point annotations. Our network introduces general local features, representing global factors from different neighborhoods based on their order positions. Then, we share query point weights to local features through point attention to reinforce impacts, which are essential in determining sparse point labels. Geometric encoding is introduced to balance query point impact and remind point position during training. As a result, one point in specific local areas can obtain global features from corresponding ones in other neighborhoods and reinforce from its query points. Experimental results on benchmark large-scale point clouds demonstrate our proposed network's state-of-the-art performance.