• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.9
• /
• pp.3782-3796
• /
• 2020

A three-dimensional (3D) reconstruction is an important research area in computer vision. The ability to detect and match features across multiple views of a scene is a critical initial step. The tracking matrix W obtained from a 3D reconstruction can be applied to structure from motion (SFM) algorithms for 3D modeling. We often fail to generate an acceptable number of features when processing face or medical images because such images typically contain large homogeneous regions with minimal variation in intensity. In this study, we seek to locate sufficient matching points not only in general images but also in face and medical images, where it is difficult to determine the feature points. The algorithm is implemented on an adaptive threshold value, a scale invariant feature transform (SIFT), affine SIFT, speeded up robust features (SURF), and affine SURF. By applying the algorithm to face and general images and studying the geometric errors, we can achieve quasi-dense matching points that satisfy well-functioning geometric constraints. We also demonstrate a 3D reconstruction with a respectable performance by applying a column space fitting algorithm, which is an SFM algorithm.

• Journal of Ubiquitous Convergence Technology
• /
• v.1 no.1
• /
• pp.9-17
• /
• 2007

Feature based stereo matching is an effective way to perform 3D building reconstruction. However, in urban scene, the cluttered background and various building structures may interfere with the performance of building reconstruction. In this paper, we propose a novel method to robustly reconstruct buildings on the basis of rectangle regions. Firstly, we propose a multi-scale linear feature detector to obtain the salient line segments on the object contours. Secondly, candidate rectangle regions are extracted from the salient line segments based on their local information. Thirdly, stereo matching is performed with the list of matching line segments, which are boundary edges of the corresponding rectangles from the left and right image. Experimental results demonstrate that the proposed method can achieve better accuracy on the reconstructed result than pixel-level stereo matching.

• 제어로봇시스템학회:학술대회논문집
• /
• 2004.08a
• /
• pp.1458-1463
• /
• 2004

Tracking is one of the most important pre-required task for many application such as human-computer interaction through gesture and face recognition, motion analysis, visual servoing, augment reality, industrial assembly and robot obstacle avoidance. Recently, 3D information of object is required in realtime for many aforementioned applications. 3D tracking is difficult problem to solve because during the image formation process of the camera, explicit 3D information about objects in the scene is lost. Recently, many vision system use stereo camera especially for 3D tracking. The 3D feature based tracking(3DFBT) which is on of the 3D tracking system using stereo vision have many advantage compare to other tracking methods. If we assumed the correspondence problem which is one of the subproblem of 3DFBT is solved, the accuracy of tracking depends on the accuracy of camera calibration. However, The existing calibration method based on accurate camera model so that modelling error and weakness to lens distortion are embedded. Therefore, this thesis proposes 3D feature based tracking method using SVM which is used to solve reconstruction problem.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.11a
• /
• pp.193-195
• /
• 2022

본 논문에서는 3차원 스캔 없이 이미지 입력만을 사용한 개인 체형을 고려한 모바일 가상 착용 시스템의 전체 과정을 설계하고 개발하였다. 이를 위하여 이미지상 인물의 자세와 체형의 추정을 통하여 3차원 인체모델(SMPL)을 추정하는 최근의 방식을 이용하였고, 앞 뒷면 의상 이미지를 2차원 texture 매핑과 평면 triangle mesh로 복원하고 의상 봉제 (sewing) 시뮬레이션을 사용하여 3차원 의상 모델을 생성하는 방법을 새롭게 개발하였다. 또한 이를 활용한 3차원 개인화된 가상 착용 모바일 앱과 서비스를 Flask와 iOS 환경에서 SceneKit을 활용하여 개발하였다. 이를 통하여 단순히 의상의 매칭과 스타일 뿐 아니라 사이즈에 따른 착용 Fit을 구매 전에 확인할 수 있는 전체 서비스를 실현 및 검증하였다.

• Advances in robotics research
• /
• v.1 no.2
• /
• pp.127-140
• /
• 2014

This work presents an IR-based system for parking assistance and obstacle detection in the automotive field that employs the Microsoft Kinect camera for fast 3D point cloud reconstruction. In contrast to previous research that attempts to explicitly identify obstacles, the proposed system aims to detect "reachable regions" of the environment, i.e., those regions where the vehicle can drive to from its current position. A user-friendly 2D traversability grid of cells is generated and used as a visual aid for parking assistance. Given a raw 3D point cloud, first each point is mapped into individual cells, then, the elevation information is used within a graph-based algorithm to label a given cell as traversable or non-traversable. Following this rationale, positive and negative obstacles, as well as unknown regions can be implicitly detected. Additionally, no flat-world assumption is required. Experimental results, obtained from the system in typical parking scenarios, are presented showing its effectiveness for scene interpretation and detection of several types of obstacle.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.06a
• /
• pp.4-7
• /
• 2022

Recently, multi-view depth estimation methods using deep learning network for the 3D scene reconstruction have gained lots of attention. Multi-view video contents have various characteristics according to their camera composition, environment, and setting. It is important to understand these characteristics and apply the proper depth estimation methods for high-quality 3D reconstruction tasks. The camera setting represents the physical distance which is called baseline, between each camera viewpoint. Our proposed methods focus on deciding the appropriate depth estimation methodologies according to the characteristics of multi-view video contents. Some limitations were found from the empirical results when the existing multi-view depth estimation methods were applied to a divergent or large baseline dataset. Therefore, we verified the necessity of obtaining the proper number of source views and the application of the source view selection algorithm suitable for each dataset's capturing environment. In conclusion, when implementing a deep learning-based depth estimation network for 3D scene reconstruction, the results of this study can be used as a guideline for finding adaptive depth estimation methods.

• Journal of Broadcast Engineering
• /
• v.27 no.4
• /
• pp.487-498
• /
• 2022

Depth map is an image that contains distance information in 3D space on a 2D plane and is used in various 3D vision tasks. Many existing depth estimation studies mainly use narrow FoV images, in which a significant portion of the entire scene is lost. In this paper, we propose a technique for generating 360° omnidirectional RGBD images from a sparse set of narrow FoV images. The proposed generative adversarial network based image generation model estimates the relative FoV for the entire panoramic image from a small number of non-overlapping images and produces a 360° RGB and depth image simultaneously. In addition, it shows improved performance by configuring a network reflecting the spherical characteristics of the 360° image.

• The KIPS Transactions:PartB
• /
• v.10B no.5
• /
• pp.515-520
• /
• 2003

This paper proposes a calibration method from two images. Camera calibration is necessarily required to obtain 3D Information from 2D images. Previous works to accomplish the camera calibration needed the calibration object or required more than three images to calculate the Kruppa equation, however, we use the geometric constraints of parallelism and orthogonality can be easily presented in man-made scenes. The task of it is to obtain intrinsic and extrinsic camera parameters. The intrinsic parameters are evaluated from vanishing points and then the extrinsic parameters which are consisted of rotation matrix and translation vector of the camera are estimated from corresponding points of two views. From the calibrated parameters, we can recover the projection matrices for each view point. These projection matrices are used to recover 3D information of the scene and can be used to visualize new viewpoints.

• Journal of KIISE:Computing Practices and Letters
• /
• v.7 no.2
• /
• pp.132-140
• /
• 2001

This paper proposes an efficient walktlu-ough animation from two images of the same scene. To make animation easily and fast, Tour Into the Picture(TIP) enables walkthrough animation from single image but lacks the reality of its foreground object when the viewpoint moves from side to side, and view morphing uses only 2D transition between two images but restricts its camera path on the line between two views. By combining advantages of these two image-based techniques, this paper suggests a new virtual navigation technique which enable natural scene transformation when the viewpoint changes in the side-to-side direction as well as in the depth direction. In our method, view morphing is employed only in foreground objects , and background scene which is perceived carelessly is mapped into cube-like 3D model as in TIP, so as to save laborious 3D reconstruction costs and improve visual realism simultaneously. To do this, we newly define a camera transformation between two images from the relationship of the spidery mesh transformation and its corresponding 3D view change. The result animation shows that our method creates a realistic 3D virtual navigation using a simple interface.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2014.11a
• /
• pp.817-820
• /
• 2014

본 논문에서는 휴대용 카메라를 이용한 3차원 공간 재구성을 위해 카메라의 실시간 포즈를 정확히 추정할 수 있는 카메라 포즈 최적화 시스템을 제안한다. 본 시스템에서는 3차원 공간에서 6차원 자유도를 가지고 움직이는 카메라의 주행 거리와 추정 포즈들 사이의 관계를 3차원 포즈 그래프로 나타냈다. 그리고 이 포즈 그래프에 대표적인 포즈 SLAM 알고리즘인 g2o를 적용함으로써, 최적화된 카메라 포즈들을 계산해낸다. 본 논문에서는 TUM 대학의 벤치마크 데이터 집합을 이용해 다양한 성능 평가 실험들을 수행하였고, 이를 통해 본 논문에서 제안한 카메라 포즈 최적화 시스템의 높은 성능을 확인할 수 있었다.