http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
윤기민(Kimin Yun),윤상두(Sangdoo Yun),최진영(Jinn Young Choi) 대한전자공학회 2015 대한전자공학회 학술대회 Vol.2015 No.6
In this paper, we propose a group violence detection framework considering motion interaction between objects. Unlike previous works, our method do not need precise object information. We use a field-like interaction feature, and build a normal model through sparsity based learning. Additionally, we measure the continuity of interaction feature field to improve the detection performance. In experiments, our method outperforms the state-of-the-art methods through qualitative and quantitative results.
비디오 안정화를 위한 bit-plane 상에서의 강인한 국소 패치 매칭 기법
김수완(Soo Wan Kim),윤기민(Kimin Yun),이현진(Hyun-Jin Lee),최진영(Jin Young Choi) 대한전자공학회 2010 대한전자공학회 학술대회 Vol.2010 No.6
In this paper, we propose robust and fast video stabilization algorithm by novel patch matching algorithm. In every frame, patches for finding camera motion are generated to contain edges and matched in the edge map from gray coded bit plane. Each individual patch results is combined to find the camera motion. After that, we keep reliable patches and discard incorrect patches for stabilization in the next frame. We showed the experimental results compared to the conventional approaches.
멀티태스크 학습을 활용한 실세계 쓰러진 사람 탐지 기술 개발
부원국(Wonkuk Boo),배강민(Kangmin Bae),윤기민(Kimin Yun),배유석(Yuseok Bae) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
Recent advancements in deep learning have stimulated the development of various datasets pertinent to human understanding. However, only a fraction of these datasets address a range of social issues, such as fallen person detection, while the majority focus on providing human keypoints and action labels. Therefore, this paper proposes a multi-task learning approach that jointly trains annotations, including keypoints and state recognitions, to enhance fallen person detection. We offer a statistical overview of human state datasets and propose methods to reconcile discrepancies in human state labels derived from multiple domains. Additionally, we present both qualitative and quantitative results of fallen person detection using benchmark datasets.