• 로봇학회논문지
• /
• 제13권1호
• /
• pp.8-15
• /
• 2018

This paper proposes a convolutional neural network model for distinguishing areas occupied by obstacles from a LiDAR image converted from a 3D point cloud. The channels of a LiDAR image used as input consist of the distances to 3D points, the reflectivities of 3D points, and the heights of 3D points from the ground. The proposed model uses a LiDAR image as an input and outputs a result of a segmented LiDAR image. The proposed model adopts refinement modules with skip connections to segment a LiDAR image. The refinement modules with skip connections in the proposed model make it possible to construct a complex structure with a small number of parameters than a convolutional neural network model with a linear structure. Using the proposed model, it is possible to distinguish areas in a LiDAR image occupied by obstacles such as vehicles, pedestrians, and bicyclists. The proposed model can be applied to recognize surrounding obstacles and to search for safe paths.

• 한국정보전자통신기술학회논문지
• /
• 제17권4호
• /
• pp.275-287
• /
• 2024

과거부터 꾸준히 지식 그래프(Knowledge Graph)와 언어 모델(LM, Language Model)의 통합에 대한 많은 연구가 다뤄지고 있다. 그 중, 지식 그래프의 구조화된 지식을 이용해 자동 텍스트 생성을 다루는 연구는 그리 활성화되지 않았다. 본 연구에서는 기존 논문들과 비슷한 수준의 특정 도메인 관련 연구 항목(Related Work)을 자동 생성하기 위한 방법론, 즉, '1) 최적의 Prompt 선정, 2) 4단계 정제기법을 통해 Triple 추출, 3) 지식 그래프 구축, 4) 관련 연구 자동 생성'을 제안한다. 제안된 방법론은 대규모 언어 모델(LLM) 중, GPT-4를 활용하고, 4단계 정제 기법을 적용하여 관련 연구를 자동으로 생성하도록 설계했다. 그렇게 설계된 모델은 Triple 추출에서 #Supp, #Cont, Fluency에서 17.3, 14.1, 4.2의 성능과 GPT-4 자동 평가 기준, 100점 기준 정제 전, 88.5점에서 정제 후, 96.5점으로 기존 논문과 비슷한 수준의 유의미한 관련 연구 자동 생성 능력을 보였다.

• 한국자기공명학회논문지
• /
• 제26권1호
• /
• pp.1-9
• /
• 2022

There are two distinct approaches to improving the quality of protein NMR structures during refinement: all-atom force fields and accumulated knowledge-assisted methods that include Rosetta. Mao et al. reported that, for 40 proteins, Rosetta increased the accuracies of their NMR-determined structures with respect to the X-ray crystal structures (Mao et al., J. Am. Chem. Soc. 136, 1893 (2014)). In this study, we calculated 32 structures of those studied by Mao et al. using all-atom force field and implicit solvent model, and we compared the results with those obtained from Rosetta. For a single protein, using only the experimental NOE-derived distances and backbone torsion angle restraints, 20 of the lowest energy structures were extracted as an ensemble from 100 generated structures. Restrained simulated annealing by molecular dynamics simulation searched conformational spaces with a total time step of 1-ns. The use of GPU-accelerated AMBER code allowed the calculations to be completed in hours using a single GPU computer-even for proteins larger than 20 kDa. Remarkably, statistical analyses indicated that the structures determined in this way showed overall higher accuracies to their X-ray structures compared to those refined by Rosetta (p-value < 0.01). Our data demonstrate the capability of sophisticated atomistic force fields in refining NMR structures, particularly when they are coupled with the latest GPU-based calculations. The straightforwardness of the protocol allows its use to be extended to all NMR structures.

• 대한임베디드공학회논문지
• /
• 제18권5호
• /
• pp.259-265
• /
• 2023

Pedestrian detection is used to effectively detect pedestrians in various situations based on deep learning. Pedestrian detection has difficulty detecting pedestrians due to problems such as camera performance, pedestrian description, height, and occlusion. Even in the same pedestrian, performance in detecting them can differ according to the height of the pedestrian. The height of general pedestrians encompasses various scales, such as those of infants, adolescents, and adults, so when the model is applied to one group, the extraction of data becomes inaccurate. Therefore, this study proposed a pedestrian detection method that fine-tunes the pedestrian area by Refining Layer and Feature Concatenation to consider various heights of pedestrians. Through this, the score and location value for the pedestrian area were finely adjusted. Experiments on four types of test data demonstrate that the proposed model achieves 2-5% higher average precision (AP) compared to Faster R-CNN and DRPN.

• Steel and Composite Structures
• /
• 제18권4호
• /
• pp.853-871
• /
• 2015

Up to now, Japan has more than 200 corrugated steel web composite beam bridges which are under construction and have been constructed, and China has more than 30 corrugated steel web composite beam bridges. The bridge type includes the simply supported beam, continuous beam, continuous rigid frame and cable stayed bridge etc. The section form has developed to the single box and multi-cell box girder from the original single box and single chamber. From the stress performance and cost saving, the span range of 50~150 m is the most competitive. At present, the design mostly adopts the computational analytical method combining the spatial bar system model, plane beam grillage model and solid model. However, the spatial bar system model is short of the refinement analysis on the space effect, such as the shear lag effect, effective distribution width problem, and eccentric load factor problem etc. Due to the similarity of the plane beam grillage method in the equivalence principle, it cannot accurately reflect the shearing stress distribution and local stress of the top and bottom plates of the box type composite beam. The solid model is very difficult to combine with the overall calculation. Moreover, the spatial grid model can achieve the refinement analysis, with the integrity of the analysis and the comprehensiveness of the stress checking calculation, and can make up the deficiency of the analytical method currently. Through the example verification of the solid model and spatial grid model, it can be seen that the calculation results for the stress and the displacement of two models are almost consistent, indicating the applicability and precision of the spatial grid model.

• 디지털콘텐츠학회 논문지
• /
• 제12권2호
• /
• pp.185-194
• /
• 2011

본 논문은 게임과 가상현실 시스템을 위한 스테이트챠트(Statecharts) 기반의 점진적인(progressive) 행위 LOD 모델을 소개하고 있다. 시스템의 리소스를 최대한 활용할 수 있도록 상황에 따라 임의의 복잡도를 가진 행위 LOD를 자유롭게 생성할 수 있는 스테이트챠트 기반의 명세 프로세스를 포함하여, 모델간 개량(refinement) 연산, 모델 간 전환(switching) 법칙, LOD 선택 정책 등을 정의하였다. 더불어, 제안된 접근방법의 가능성을 보여주기 위해, 가상 자동차가 단계적으로 설계되어가면서 점진적인 LOD 모델이 되는 예를 들고 있다.

• 전자공학회논문지T
• /
• 제36T권1호
• /
• pp.93-102
• /
• 1999

본 논문은 내용 기반 비디오 질의를 위한 데이터 모델과 검색 방법을 제안한다. 하나의 비디오를 같은 길이의 프레임(frame)들의 집합 즉 비디오-윈도우로 나눈 후에 각각의 비디오-윈도우를 다차원 공간의 한 점으로 사상시킨다. 인접한 비디오-윈도우를 연결하면 하나의 비디오는 다차원 공간에서의 하나의 궤적(trajectory)이된다. 두 비디오-윈도우의 유서성은 두 점의 유클리디안 거리로 정의되며, 비디오 단편(segment)의 유사성 비교는 궤적을 비교함으로써 검사한다. 여과(filtering), 정제(refinement)과정을 가지는 새로운 검색 방법을 개발한다. 새로운 검색 방법을 여과/정제 과정이 없는 질의 결과가 정확하고, 질의 처리 속도는 약 4.7배 향상되었다.

• 대한기계학회논문집A
• /
• 제25권12호
• /
• pp.1926-1932
• /
• 2001

In this second part of the paper, the automatic mesh generation and remeshing algorithm using bubble packing method is applied to the nonlinear problem. The remeshing/refinement procedure is necessary in the large deformation process especially because the mesh distortion deteriorates the convergence and accuracy. To perform the nonliear analysis, the transfer of state variables such as displacement and strain is added to the algorithm of Part 1. The equilibrium equation based on total Lagrangian formulation and elasto-viscoplastic model is used. For the numerical experiment, the upsetting process including the contact constraint condition is analyzed by two refinement criteria. And from the result, it is addressed that the present algorithm can generate the refined meshes easily at the largely deformed area with high error.

• 한국전산구조공학회:학술대회논문집
• /
• 한국전산구조공학회 2002년도 가을 학술발표회 논문집
• /
• pp.615-622
• /
• 2002

In this paper, the modification of double projection method for the adaptive analysis of Element-free Galerkin(EFG) method were proposed. As results of the double projection method, the smoothed error profile that is adequate for adaptive analysis was obtained by re-projection of error that means the differences of EFG stress and projected stress. However, it was found that the efficiency of double projection method is degraded as increase of the numerical integration order. Since, the iterative refinement to single step error estimation made the same effect as increasing of integration order, the application of the iterative refinement base on double projection method could be produced the inadequately refined analysis model. To overcome this defect, a modified scheme of double projection were proposed. In the numerical example, the results did not show degradation of double projection effect in iterative refinement and the efficiency of proposed scheme were proved.

• The Korean Journal of Ceramics
• /
• 제5권4호
• /
• pp.341-347
• /
• 1999

An ambiguity on the correct room temperature structure of $Bi_5Ti_3FeO_{15}$ was resolved using a combined Rietveld refinement of neutron and X-ray diffraction. The structure of this compound has been reported to have a space group of F2mm (adopting 2-fold rotation symmetry along the c-axis) or A21am. However, our diffraction, study reveals that some reflections would violate F-centering and confirm that the belong to $A2_1$am. Out refinement with the space group of $A2_1$am converged at $R_p=6.85%, R_wp=9.23%$ and $\chi^2$=1.66 for an isotropic temperature model with 85 variables. The lattice constants are a=5.4677(1) $\AA$, b=5.4396(1) $\AA$, and c=41.2475(8)$\AA$. In structure, Ti/Fe atoms at the oxygen octahedral sites of the perovskite unit are completely disordered, resulting in that these atoms are transparent in neutron diffraction. The octahedra of the perovskite unit are relatively displaced along the a-axis against the Bi atoms, which contribute as a major component to the spontaneous polarization of $Bi_5Ti_3FeO_{15}$.