• Journal of Broadcast Engineering
• /
• v.27 no.2
• /
• pp.174-184
• /
• 2022

3D human pose estimation models can be classified into a multi-view model and a single-view model. In general, the multi-view model shows superior pose estimation performance compared to the single-view model. In the case of the single-view model, the improvement of the 3D pose estimation performance requires a large amount of training data. However, it is not easy to obtain annotations for training 3D pose estimation models. To address this problem, we propose a method to generate pseudo ground-truths of multi-view human pose data from a multi-view model and exploit the resultant pseudo ground-truths to train a single-view model. In addition, we propose a multi-view consistency loss function that considers the consistency of poses estimated from multi-view images, showing that the proposed loss helps the effective training of single-view models. Experiments using Human3.6M and MPI-INF-3DHP datasets show that the proposed method is effective for training single-view 3D human pose estimation models.

• Journal of Cadastre & Land InformatiX
• /
• v.51 no.1
• /
• pp.67-77
• /
• 2021

Spatial data is becoming very important as a medium that connects various data produced in smart cities, digital twins, autonomous driving, smart construction, and other applications. In addition, the rapid construction and update of spatial information is becoming a hot topic to satisfy the diverse needs of consumers in this field. This study developed a software prototype that can match the position of an image-based 3D building model produced without Ground Control Points using existing spatial data. As a result of applying this software to the test area, the 3D building model produced based on the image and the existing spatial data show a high positional matching rate, so that it can be widely used in applications requiring the latest 3D spatial data.

• The KIPS Transactions:PartB
• /
• v.9B no.2
• /
• pp.255-262
• /
• 2002

The Virtual Reality (VR) is an appealing subject which can be applied to various areas because of its merit - removal of time limits and space. Recently, as the technology of xDSL spreads widely, a concern of VR is on the on-line service of 3D model data in real time. But, the immensity of 3D model is an obstacle to achieve these endeavors. To solve these problems, the image based VR technique is applied. The proposed method in this paper is one of solutions on the immensity problem of 3D model data in the on-line services. This paper exploits the mixed technique of image based VR and surface rendering based on volume rendering. By using the proposed method, we can solve the immensity problem. Consequently, tole service user can explore virtual 2D volume model with almost equal to reality of 3D volume model. Furthermore, this paper explains a method to implement this service on general web environments. Of course, to fulfill these procedures, additional skills which reduce consuming time in data mining are also mentioned. The contribution of this paper is to provide a practical method for handling of large volume data web service in real-time. illustrative examples are presented to show the effectiveness of the proposed method.

• Korean Journal of Remote Sensing
• /
• v.32 no.5
• /
• pp.511-527
• /
• 2016

We propose a rotating stereo frame camera system to acquire indoor images with a low cost. For experiments, we selected a test site and acquired images using the proposed system and control points using a total station. Using these data, we generated various indoor 3D models using commercial photogrammetric software, PhotoScan. We then performed qualitative and quantitative analysis of the generated indoor 3D models to investigate the possibility of the indoor modeling using the proposed system. From the results, it is confirmed that the generated indoor models using the proposed system can be applicable to the services not inquiring high accuracy.

• 한국HCI학회:학술대회논문집
• /
• 2006.02a
• /
• pp.1387-1392
• /
• 2006

본 논문에서는 SVM (Support Vector Machine)을 기반으로 하여 인체의 뇌 하부구조인 해마에 대한 지능적 형상분석 방법을 제공한다. 일반적으로 의료 영상으로부터 해마의 형상 분석을 하기 위해서는 충분한 임상 데이터를 필요로 한다. 하지만 현실적으로 많은 양의 표본들을 얻는 것이 쉽지 않기 때문에 전문가의 지식을 기반으로 한 작업이 수반되어야 한다. 결국 이러한 요소들이 분석 작업을 어렵게 한다. 의학 기술이 복잡해 지면서 최근의 형상 분석 연구는 점차 통계적 모델을 기반으로 진행되고 있다. 본 연구에서는 해마로부터 고해상도의 매개변수형 모델을 만들어 형상 표현으로 이용하고, 집단간 분류 작업에 SVM 알고리즘을 적용하는 지능적 분석 방법을 구현한다. 우선 메쉬 데이터로부터 물리변형모델 기반의 매개변수 모델을 구축하고, PDM (point distribution model) 방법을 적용하여 두 집단을 대표하는 평균 모델을 생성한다. 마지막으로 SVM 기반의 이진 분류기를 구축하여 집단간 분류 작업을 수행한다. 구현한 모델링 방법과 분류기의 성능을 평가하기 위하여 본 연구에서는 네 가지 커널 함수 (linear, radial basis function, polynomial, sigmoid)들을 적용한다. 본 논문에서 제시한 매개변수형 모델은 다양한 형태의 의료 데이터로부터 보편적인 3차원 모델을 생성하고, 또한 모델의 전역적, 국부적인 특징들을 복합적으로 표현할 수 있기 때문에 통계적 형상분석에 적합하다. 그리고 SVM 기반의 분류기는 적은 수의 학습 데이터로부터 정상인 해마 집단과 간질 환자 집단간의 정확한 분류를 가능하게 한다.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.10 no.1
• /
• pp.1-9
• /
• 2007

3D data are in great demand for pedestrian navigation recently. For pedestrian navigation, we needs to reconstruct 3D model in detail from people's eye. In order to present spatial features in detail for pedestrian navigation, it is indispensable to develop 3D model not only in outdoor environment but also in indoor environment such as underground shopping complex. However, it is very difficult to acquire 3D data efficiently by mobile mapping without GPS. In this research, 3D shape was acquired by Laser scanner, and texture by CCD(Charge Coupled Device) sensor. Continuous changes position and attitude of sensors were measured by IMU(Inertial Measurement Unit). Moreover, IMU was corrected by relative orientation of CCD images without GPS(Global Positioning System). In conclusion, Reliable, quick, and handy method for acquiring 3D data for indoor environment is proposed by a combination of a digital camera and a laser scanner with IMU.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.34 no.5
• /
• pp.479-493
• /
• 2016

LiDAR(Light Detection and Ranging) data acquired from ALS(Airborne Laser Scanner) has been intensively utilized to reconstruct object models. Especially, researches for 3D modeling from LiDAR data have been performed to establish high quality spatial information such as precise 3D city models and true orthoimages efficiently. To reconstruct object models from irregularly distributed LiDAR point clouds, sensor calibration, noise removal, filtering to separate objects from ground surfaces are required as pre-processing. Classification and segmentation based on geometric homogeneity of the features, grouping and representation of the segmented surfaces, topological analysis of the surface patches for modeling, and accuracy assessment are accompanied by modeling procedure. While many modeling methods are based on the segmentation process, this paper proposed to extract key points directly for building modeling without segmentation. The method was applied to simulated and real data sets with various roof shapes. The results demonstrate feasibility of the proposed method through the accuracy analysis.

• Journal of Broadcast Engineering
• /
• v.28 no.3
• /
• pp.329-332
• /
• 2023

3D Object Detection using deep neural network has been developed a lot for obstacle detection in autonomous vehicles because it can recognize not only the class of target object but also the distance from the object. But in the case of 3D Object Detection models, the detection performance for distant objects is lower than that for nearby objects, which is a critical issue for autonomous vehicles. In this paper, we introduce a technique that increases the performance of 3D object detection models, particularly in recognizing distant objects, by generating virtual 3D vehicle data and adding it to the dataset used for model training. We used a spherical point tracing method that leverages the characteristics of 3D LiDAR sensor data to create virtual vehicles that closely resemble real ones, and we demonstrated the validity of the virtual data by using it to improve recognition performance for objects at all distances in model training.

• Electronics and Telecommunications Trends
• /
• v.12 no.6 s.48
• /
• pp.15-23
• /
• 1997

본 논문에서는 실존하는 물체의 3차원 모델을 만드는 방법들 중에서 물체로부터 직접 정보를 얻어내어 매우 사실적인 3차원 모델을 제작하는 기법들에 대한 현재까지의 기술들에 대해 분석해보고 앞으로의 발전 방향에 대한 전망을 해보았다. 실제 물체의 데이터를 얻어내는 방법은 크게 세가지로 나뉘며 3차원 디지타이저, 3차원 레이저 스캐너 그리고 사진을 이용하는 방법이 있다. 이러한 방법들이 기초적인 요구 사항을 만족하고 있는지와 핵심기술은 어떠한 것들인지를 분석해본다.

• 전자공학회논문지 IE
• /
• v.46 no.1
• /
• pp.37-43
• /
• 2009

In this paper, we propose a new algorithm for 3D model retrieval based on spherical coordinate system. We obtains sample points in a polygons on 3D model. We convert a point in cartesian coordinates(x, y, z) to it in spherical coordinate. 3D shape features are achieved by adopting distribution of zenith of sample point in spherical coordinate. We used Osada's method for obtaining sample points on 3D model and the PCA method for the pose standardization 3D model. Princeton university's benchmark data was used for this research. Experimental results numerically show the precision improvement of proposed algorithm 12.6% in comparison with Vranic's depth buffer-based feature vector algorithm.