• Journal of the Korea Computer Graphics Society
• /
• v.21 no.5
• /
• pp.37-44
• /
• 2015

This paper presents a novel 3D model representation, called hybrid model representation, to overcome existing 3D volume-based indoor scene reconstruction mechanism. In indoor 3D scene reconstruction, volume-based model representation can reconstruct detailed 3D model for the narrow scene. However it cannot reconstruct large-scale indoor scene due to its memory consumption. This paper presents a memory efficient plane-hash model representation to enlarge the scalability of the indoor scene reconstruction. Also, the proposed method uses plane-hash model representation to reconstruct large, structural planar objects, and at the same time it uses volume-based model representation to recover small detailed region. Proposed method can be implemented in GPU to accelerate the computation and reconstruct the indoor scene in real-time.

• 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 the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.25 no.6_1
• /
• pp.545-555
• /
• 2007

3D buildings are the most prominent feature comprising urban scene. A few of mega-cities in the globe are virtually reconstructed in photo-realistic 3D models, which becomes accessible by the public through the state-of-the-art online mapping services. A lot of research efforts have been made to develop automatic reconstruction technique of large-scale 3D building models from remotely sensed data. However, existing methods still produce irregular building polygons due to errors induced partly by uncalibrated sensor system, scene complexity and partly inappropriate sensor resolution to observed object scales. Thus, a geometric regularization technique is urgently required to rectify such irregular building polygons that are quickly captured from low sensory data. This paper aims to develop a new method for regularizing noise building outlines extracted from airborne LiDAR data, and to evaluate its performance in comparison with existing methods. These include Douglas-Peucker's polyline simplication, total least-squared adjustment, model hypothesis-verification, and rule-based rectification. Based on Minimum Description Length (MDL) principal, a new objective function, Geometric Minimum Description Length (GMDL), to regularize geometric noises is introduced to enhance the repetition of identical line directionality, regular angle transition and to minimize the number of vertices used. After generating hypothetical regularized models, a global optimum of the geometric regularity is achieved by verifying the entire solution space. A comparative evaluation of the proposed geometric regulator is conducted using both simulated and real building vectors with various levels of noise. The results show that the GMDL outperforms the selected existing algorithms at the most of noise levels.