• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.660-662
• /
• 2023

3차원 포인트 클라우드의 의미적 분할은 환경을 구성하는 물체 단위로 포인트 클라우드를 분할하는 작업으로서, 환경의 3차원적 구성을 이해하고 환경과 상호작용에 필수적인 시각 지능을 요구한다. 본 논문에서는 포인트 클라우드에서 추출하는 3차원 기하학적 특징과 함께 멀티-뷰 영상에서 추출하는 2차원 시각적 특징들도 활용하는 새로운 3차원 포인트 클라우드 의미적 분할 모델 MFNet을 제안한다. 제안 모델은 서로 이질적인 2차원 시각적 특징과 3차원 기하학적 특징의 효과적인 융합을 위해, 새로운 중기 융합 전략과 멀티-모달 교차 주의집중을 이용한다. 본 논문에서는 ScanNetV2 벤치마크 데이터 집합을 이용한 다양한 실험들을 통해, 제안 모델 MFNet의 우수성을 입증한다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.06a
• /
• pp.829-831
• /
• 2022

포인트 클라우드 콘텐츠는 움직임이 있는 콘텐츠를 연속된 프레임에 3 차원 위치정보와 대응하는 색상으로 기록한 데이터이다. 강체 포인트 클라우드 데이터를 정합하기 위해서는 고전적인 방법이지만 강력한 ICP 정합 알고리즘을 사용한다. 그러나 국소적인 모션 벡터가 있는 비 강체 포인트 클라우드 콘텐츠는 기존의 ICP 정합 알고리즘을 통해서는 프레임 간 정합이 불가능하다. 본 논문에서는 비 강체 포인트 클라우드 콘텐츠를 지역적 확률 모델을 사용하여 프레임 간 포인트의 쌍을 맺고 개별 포인트 간의 모션벡터를 구해 정합 하는 방법을 제안한다. 정합 대상의 데이터를 2 차원 투영을 하여 구조화시키고 정합 할 데이터를 투영하여 후보군 포인트를 선별한다. 선별된 포인트에서 깊이 값 비교와 좌표 및 색상 유사도를 측정하여 적절한 쌍을 찾아준다. 쌍을 찾은 후 쌍으로 모션 벡터를 더하여 정합을 수행하면 비 강체 포인트 클라우드 콘텐츠 데이터에 대해서도 정합이 가능해진다.

• Korean Journal of Remote Sensing
• /
• v.36 no.6_1
• /
• pp.1349-1365
• /
• 2020

In recent years, in the fields of smart cities and digital twins, research on model generation is increasing due to the advantage of acquiring actual 3D coordinates by using point clouds. In addition, there is an increasing demand for a solid model that can easily modify the shape and texture of the building. In this paper, we propose a method to create a clustered solid building model based on point cloud data. The proposed method consists of five steps. Accordingly, in this paper, we propose a method to create a clustered solid building model based on point cloud data. The proposed method consists of five steps. In the first step, the ground points were removed through the planarity analysis of the point cloud. In the second step, building area was extracted from the ground removed point cloud. In the third step, detailed structural area of the buildings was extracted. In the fourth step, the shape of 3D building models with 3D coordinate information added to the extracted area was created. In the last step, a 3D building solid model was created by giving texture to the building model shape. In order to verify the proposed method, we experimented using point clouds extracted from unmanned aerial vehicle images using commercial software. As a result, 3D building shapes with a position error of about 1m compared to the point cloud was created for all buildings with a certain height or higher. In addition, it was confirmed that 3D models on which texturing was performed having a resolution of less than twice the resolution of the original image was generated.

• Korean Journal of Remote Sensing
• /
• v.34 no.6_1
• /
• pp.941-951
• /
• 2018

Utilization and demand of UAV (unmanned aerial vehicle) for the generation of 3D city model are increasing. In this study, we performed an experiment to adjustment position/orientation of UAV with incomplete attitude information and to extract point cloud data. In order to correct the attitude of the UAV, the rotation angle was calculated by using the continuous position information of UAV movements. Based on this, the corrected position/orientation information was obtained by applying IBA (Incremental Bundle Adjustment) based on photogrammetry. Each pair was transformed into an epipolar image, and the MDR (Multi-Dimensional Relaxation) technique was applied to obtain high precision DSM. Each extracted pair is aggregated and output in the form of a single point cloud or DSM. Using the DJI inspire1 and Phantom4 images, we can confirm that the point cloud can be extracted which expresses the railing of the building clearly. In the future, research will be conducted on improving the matching performance and establishing sensor models of oblique images. After that, we will continue the image processing technology for the generation of the 3D city model through the study of the extraction of 3D cloud It should be developed.

• Korean Journal of Remote Sensing
• /
• v.35 no.6_1
• /
• pp.973-985
• /
• 2019

In this paper, we propose a method to separate the ground and building areas and generate building models automatically through planarity analysis using UAV (Unmanned Aerial Vehicle) based point cloud. In this study, proposed method includes five steps. In the first step, the planes of the point cloud were extracted by analyzing the planarity of the input point cloud. In the second step, the extracted planes were analyzed to find a plane corresponding to the ground surface. Then, the points corresponding to the plane were removed from the point cloud. In the third step, we generate ortho-projected image from the point cloud ground surface removed. In the fourth step, the outline of each object was extracted from the ortho-projected image. Then, the non-building area was removed using the area, area / length ratio. Finally, the building's outer points were constructed using the building's ground height and the building's height. Then, 3D building models were created. In order to verify the proposed method, we used point clouds made using the UAV images. Through experiments, we confirmed that the 3D models of the building were generated automatically.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.187-189
• /
• 2020

본 논문에서는 포인트 클라우드 정합 성능 향상을 위해 기하적 복잡도가 낮은 정점들의 영향을 최소화하는 포인트 클라우드 리샘플링 방법을 제안한다. 3 차원 특징 기술자(3D feature descriptor)를 기반으로 하는 포인트 클라우드 정합은 정점 법선 벡터의 변화량을 특징으로 사용한다. 따라서 강건한 특징은 대부분 정점 법선 벡터의 변화량이 큰 영역에서 추출된다. 반면에 정점 법선 벡터의 변화량이 거의 없는 평면 영역은 정합 수행 시에 이상점(outlier)으로 작용할 수 있으므로 해당 정점들이 정합 과정에 미치는 영향을 최소화해야 한다. 제안하는 방법은 모델 포인트 클라우드의 기하적 복잡도를 고려한 리샘플링을 통해 전체 정점의 수 대비 복잡도가 낮은 정점들의 비율을 낮추어 이상점이 정합 과정에 미치는 영향을 최소화하고 정합 성능을 향상시켰다.

• The Transactions of the Korea Information Processing Society
• /
• v.13 no.7
• /
• pp.335-347
• /
• 2024

Open-vocabulary 3D point cloud instance segmentation (OV-3DIS) is a challenging visual task to segment a 3D scene point cloud into object instances of both base and novel classes. In this paper, we propose a novel model Open3DME for OV-3DIS to address important design issues and overcome limitations of the existing approaches. First, in order to improve the quality of class-agnostic 3D masks, our model makes use of T3DIS, an advanced Transformer-based 3D point cloud instance segmentation model, as mask proposal module. Second, in order to obtain semantically text-aligned visual features of each point cloud segment, our model extracts both 2D and 3D features from the point cloud and the corresponding multi-view RGB images by using pretrained CLIP and OpenSeg encoders respectively. Last, to effectively make use of both 2D and 3D visual features of each point cloud segment during label assignment, our model adopts a unique feature ensemble method. To validate our model, we conducted both quantitative and qualitative experiments on ScanNet-V2 benchmark dataset, demonstrating significant performance gains.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.12
• /
• pp.505-518
• /
• 2023

3D point cloud semantic segmentation is a computer vision task that involves dividing the point cloud into different objects and regions by predicting the class label of each point. Existing 3D semantic segmentation models have some limitations in performing sufficient fusion of multi-modal features while ensuring both characteristics of 2D visual features extracted from RGB images and 3D geometric features extracted from point cloud. Therefore, in this paper, we propose MMCA-Net, a novel 3D semantic segmentation model using 2D-3D multi-modal features. The proposed model effectively fuses two heterogeneous 2D visual features and 3D geometric features by using an intermediate fusion strategy and a multi-modal cross attention-based fusion operation. Also, the proposed model extracts context-rich 3D geometric features from input point cloud consisting of irregularly distributed points by adopting PTv2 as 3D geometric encoder. In this paper, we conducted both quantitative and qualitative experiments with the benchmark dataset, ScanNetv2 in order to analyze the performance of the proposed model. In terms of the metric mIoU, the proposed model showed a 9.2% performance improvement over the PTv2 model using only 3D geometric features, and a 12.12% performance improvement over the MVPNet model using 2D-3D multi-modal features. As a result, we proved the effectiveness and usefulness of the proposed model.

• Journal of IKEEE
• /
• v.27 no.4
• /
• pp.486-493
• /
• 2023

In this paper, we propose an automatic hand gesture area extraction and recognition technique using FMCW radar-based point cloud and LSTM. The proposed technique has the following originality compared to existing methods. First, unlike methods that use 2D images as input vectors such as existing range-dopplers, point cloud input vectors in the form of time series are intuitive input data that can recognize movement over time that occurs in front of the radar in the form of a coordinate system. Second, because the size of the input vector is small, the deep learning model used for recognition can also be designed lightly. The implementation process of the proposed technique is as follows. Using the distance, speed, and angle information measured by the FMCW radar, a point cloud containing x, y, z coordinate format and Doppler velocity information is utilized. For the gesture area, the hand gesture area is automatically extracted by identifying the start and end points of the gesture using the Doppler point obtained through speed information. The point cloud in the form of a time series corresponding to the viewpoint of the extracted gesture area is ultimately used for learning and recognition of the LSTM deep learning model used in this paper. To evaluate the objective reliability of the proposed technique, an experiment calculating MAE with other deep learning models and an experiment calculating recognition rate with existing techniques were performed and compared. As a result of the experiment, the MAE value of the time series point cloud input vector + LSTM deep learning model was calculated to be 0.262 and the recognition rate was 97.5%. The lower the MAE and the higher the recognition rate, the better the results, proving the efficiency of the technique proposed in this paper.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.32 no.5D
• /
• pp.527-532
• /
• 2012

Terrestrial LiDAR emerges as a main mapping technology for indoor 3D cadastre, cultural heritage conservation and, building management in that it provides fast, accurate, and reliable 3D data. In this paper, a new 3D modeling method consisting of segmentation stage and outline extraction stage is proposed to develop indoor 3D model from the terrestrial LiDAR. In the segmentation process, RANSAC and a refinement grid is used to identify points that belong to identical planar planes. In the outline tracing process, a tracing grid and a data conversion method are used to extract outlines of indoor 3D models. However, despite of an improvement of productivity, the proposed approach requires an optimization process to adjust parameters such as a threshold of the RANSAC and sizes of the refinement and outline extraction grids. Furthermore, it is required to model curvilinear and rounded shape of the indoor structures.