http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
삼차원 객체 인식을 위한 라이다-카메라 후기 융합 기술
최광진(Kwangjin Choi),김지송(Jisong Kim),성민재(Minjae Seong),방건호(Geonho Bang),최준원(Jun Won Choi) 한국자동차공학회 2023 한국자동차공학회 부문종합 학술대회 Vol.2023 No.5
In this paper, we present a late-fusion method for integrating LiDAR and camera data to enhance 3D object detection in autonomous driving scenarios. Accurate environmental perception is vital for autonomous vehicles, prompting the exploration of sensor fusion methods that leverage complementary data from multiple sensors. Our work focuses on LiDAR-camera sensor fusion, aiming to merge positional data from LiDAR with semantic information from cameras. Fusion approaches are classified into early, middle, and late fusion, based on the timing of sensor data integration. We present a high-accuracy 3D object detection framework that exploits the benefits of late fusion in integrating LiDAR and camera data. We employed 3D and 2D detectors to generate feature maps and bounding boxes from each sensors. A transformer decoder is utilized to fuse LiDAR and camera data, using the queries to initialize the LiDAR feature map and 3D proposal positions. Finally, the classification scores of the 3D bounding boxes are rescaled to enhance overall accuracy. To evaluate the effectiveness of our approach, we conduct experiments on the KITTI dataset and outperformed the existing late fusion and baseline models.
Shortest Path Search Scheme with a Graph of Multiple Attributes
Jongwan Kim(김종완),KwangJin Choi(최광진),Dukshin Oh(오덕신) 한국컴퓨터정보학회 2020 韓國컴퓨터情報學會論文誌 Vol.25 No.12
그래프 이론에서 최소비용 경로는 시작 노드와 도착 노드 사이의 최단 경로를 탐색하여 구한다. 최소비용은 두 노드 사이의 거리나 가격의 차이를 1차원 값으로 계산하며 연결된 노드 사이의 최소비용의 합을 구성하는 노드와 간선이 최단 경로다. 그러나 각 노드가 다중속성을 갖는 경우에는 경로에서 나타날 수 있는 비용의 종류 또한 속성의 개수만큼이므로 최단 경로를 판단하기에는 어려움이 있다. 본 논문에서는 사용자의 다양한 요구사항을 만족할 수 있도록 유클리드 거리를 사용하여 다중속성을 반영한 최단 경로 탐색 기법을 제안한다. 실험에서는 1차원 값에 대한 최단 경로와 2차원 속성에 대한 유클리드 거리를 이용한 최단 경로가 다르게 탐색 되었다. 다중속성에서도 단일 속성과 차별화된 사용자의 선호 속성이 반영된 것으로 나타났다. 결과적으로 다중속성이 반영됨으로써 사용자의 다양한 요구사항을 만족시킬 수 있게 되었다. In graph theory, the least-cost path is discovered by searching the shortest path between a start node and destination node. The least cost is calculated as a one-dimensional value that represents the difference in distance or price between two nodes, and the nodes and edges that comprise the lowest sum of costs between the linked nodes is the shortest path. However, it is difficult to determine the shortest path if each node has multiple attributes because the number of cost types that can appear is equal to the number of attributes. In this paper, a shortest path search scheme is proposed that considers multiple attributes using the Euclidean distance to satisfy various user requirements. In simulation, we discovered that the shortest path calculated using one-dimensional values differs from that calculated using the Euclidean distance for two-dimensional attributes. The user’s preferences are reflected in multi attributes and it was different from one-dimensional attribute. Consequently, user requirements could be satisfied simultaneously by considering multiple attributes.
유영덕(Youngduk Yoo),최광진(Kwangjin Choi) 한국자동차공학회 2005 한국자동차공학회 춘 추계 학술대회 논문집 Vol.2005 No.5_2
In this Paper, We present the Analysis method using Robust Design Concept. At present Robust Design Concept used world widely in Automotive Component Design. In this case we practiced S/N Ratio Reduction in Hood assy gap variance to the extent of 2.3 ㏈ . And We could improve Hood lifting amount. We think that Many analysis method using Robust Design Concept will be developed in the future.
레이더 카메라 센서 융합을 통한 우천 및 야간 상황 3차원 물체 검출
김지송(Jisong Kim),성민재(Minjae Seong),방건호(Geonho Bang),최광진(Kwangjin Choi),최준원(Jun Won Choi) 한국자동차공학회 2023 한국자동차공학회 부문종합 학술대회 Vol.2023 No.5
In recent years, camera-based 3D object detection has garnered significant attention. However, the performance of these models tends to be compromised under challenging conditions such as rain or nighttime, owing to the inherent limitations of the sensors. In this study, we developed a radar-camera fusion model to assess the potential for enhancing object detection capabilities in adverse weather and low-light situations by incorporating radar sensors. To mitigate model overfitting, we employed suitable data preprocessing and augmentation techniques for both radar and camera inputs. Additionally, we converted the radar and camera features into a birds-eye view (BEV) representation and fused them at the feature-level. We conducted experiments using the nuScenes dataset to evaluate the models performance, and subdivided the validation dataset into four distinct scenarios to appraise the model under various conditions. The findings demonstrated that our proposed model outperformed the camera-only approach, which served as the baseline model, across all tested scenarios.
금속이온 잉크와 탄소나노튜브 용액의 에어브러싱 적층 기반 고성능 고내구성 박막형 히팅 모듈의 수월 제작 기술
김민욱(Minwook Kim),김민규(Mingyu Kim),김광준(Kwangjun Kim),박재원(Jaewon Park),이민영(Minyoung Lee),최현식(Hyunsik Choi),박종갑(Jonggab Park),노현찬(Hyeonchan Noh),윤부현(Boohyeon Youn),정은창(Eunchang Jeong),최광진(Kwangjin Choi),허 대한전기학회 2020 대한전기학회 학술대회 논문집 Vol.2020 No.7