• Proceedings of the Korean Information Science Society Conference
• /
• 2005.11a
• /
• pp.682-684
• /
• 2005

본 논문에서는 동적인 실제 객체로부터 얻어진 깊이 맵으로부터 3차원 메쉬 모델을 생성할 때, 영상의 경계정보를 기반으로 효율적인 비정규 메쉬를 생성하는 기법을 제안한다. 우선 깊이 맵으로부터 소수의 특징점과 경계영역에서의 실루엣 점을 추출한다. Delaunay 삼각화 기법을 적용할 때 경계 피부에 발생하는 불필요한 삼각형들을 효율적으로 제거하기 위해 실루엣점으로 구성된 경계 라인을 제약조건으로 사용한다. 즉 깊이 맵으로부터 경계 영역 정보를 추출하고 이를 기반으로 관심 객체의 비정규 삼각 메쉬에 존재하는 불필요한 외부 삼각형을 제거한다. 최종적으로 생성된 3차원 메쉬에 포함된 형상 노이즈를 제거하기 위해 메쉬 스무딩 기법을 적용하고, 깊이 맵과 동시에 획득된 컬러 영상을 텍스쳐링하여 3차원 메쉬를 생성한다.

• Journal of the Institute of Electronics Engineers of Korea SC
• /
• v.46 no.6
• /
• pp.28-37
• /
• 2009

In this paper, we propose an obstacle localization method using a road feature based on a V-disparity map binarized by a maximum frequency value. In a conventional method, the detection performance is severely affected by the size, number and type of obstacles. It's especially difficult to extract a large obstacle or a continuous obstacle like a median strip. So we use a road feature as a new decision standard to localize obstacles irrespective of external environments. A road feature is proper to be a new decision standard because it keeps its rough feature very well in V-disparity under environments where many obstacles exist. And first of all, we create a binary V-disparity map using a maximum frequency value to extract a road feature easily. And then we compare the binary V-disparity map with a median value to remove noises. Finally, we use a linear interpolation for rows which have no value. Comparing this road feature with each column value in disparity map, we can localize obstacles robustly. We also propose a post-processing technique to remove noises made in obstacle localization stage. The results in real road tests show that the proposed algorithm has a better performance than a conventional method.

• Journal of Korea Spatial Information System Society
• /
• v.6 no.1 s.11
• /
• pp.87-100
• /
• 2004

When spatial data which a user requests are stored various data servers in web mapping environment, it is necessary to access one more map servers to gain data and then integrate them. This paper designs and implements a fusion map server for web mapping. The proposed fusion map server has three characteristics. First, it has customizing metadata which reflect various user's requests. Second, it provides fusion services which integrate spatial and non-spatial data. Last, it uses the dynamic liking method to support the fusion services efficiently. We design and implement a publisher client, a map server and a fusion map server for proposed map server.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.22 no.10
• /
• pp.2102-2111
• /
• 1997

Hand written numeral is projected on the 3D space after pre-processing of inputs and it makes a index by tracking of numerals. It computes the distance between extracted every features. It is used by input part of recognition process from the statistical historgram of the normalization of data in order to adaptation from variation. One hundred unmeral patterns have used for making a standard feature map and 100 pattern for the recogintion experiment. The result of it, we have the recoginition rete is 93.5% based on thresholding is 0.20 and 97.5% based on 0.25.

• KIPS Transactions on Software and Data Engineering
• /
• v.5 no.11
• /
• pp.607-614
• /
• 2016

In this paper, we propose an intelligent fire learning and detection system using convolutional neural networks (CNN). Through the convolutional layer of the CNN, various features of flame and smoke images are automatically extracted, and these extracted features are learned to classify them into flame or smoke or no fire. In order to detect fire in the image, candidate fire regions are first extracted from the image and extracted candidate regions are passed through CNN. Experimental results on various image shows that our system has better performances over previous work.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2021.06a
• /
• pp.26-27
• /
• 2021

본 논문에서는 자막방송 제공을 위해 방송콘텐츠를 이해하는 방법으로 잔차 합성곱 순환신경망 기반 음향 사건 분류 기법을 제안한다. 제안된 기법은 잔차 합성곱 신경망과 순환 신경망을 연결한 구조를 갖는다. 신경망의 입력 특징으로는 멜-필터벵크 특징을 활용하고, 잔차 합성곱 신경망은 하나의 스템 블록과 5개의 잔차 합성곱 신경망으로 구성된다. 잔차 합성곱 신경망은 잔차 학습으로 구성된 합성곱 신경망과 기존의 합성곱 신경망 대비 특징맵의 표현 능력 향상을 위해 합성곱 블록 주의 모듈로 구성한다. 추출된 특징맵은 순환 신경망에 연결되고, 최종적으로 음향 사건 종류와 시간정보를 추출하는 완전연결층으로 연결되는 구조를 활용한다. 제안된 모델 훈련을 위해 라벨링되지 않는 데이터 활용이 가능한 평균 교사 모델을 기반으로 훈련하였다. 제안된 모델의 성능평가를 위해 DCASE 2020 챌린지 Task 4 데이터 셋을 활용하였으며, 성능 평가 결과 46.8%의 이벤트 단위의 F1-score를 얻을 수 있었다.

• Journal of KIISE:Software and Applications
• /
• v.29 no.6
• /
• pp.386-395
• /
• 2002

In this paper, we suggest dynamic genetic tree-maps(DGTM) using optimal features on recognizing data. The DGTM uses the genetic algorithm about the importance of features rarely considerable on conventional neural networks and introduces GTM(genetic tree-maps) using tree structure according of the priority of features. Hence, we propose the extended formula, DGTM(dynamic GTM) has dynamic functions to separate and merge the neuron of neural network along the similarity of features.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2021.06a
• /
• pp.180-182
• /
• 2021

본 논문은 눈 랜드마크 위치 검출과 시선 방향 벡터 추정이 하나의 딥러닝 네트워크로 통합된 시선 추정 네트워크를 제안한다. 제안하는 네트워크는 Stacked Hourglass Network[1]를 백본(Backbone) 구조로 이용하며, 크게 랜드마크 검출기, 특징 맵 추출기, 시선 방향 추정기라는 세 개의 부분으로 구성되어 있다. 랜드마크 검출기에서는 눈 랜드마크 50개 포인트의 좌표를 추정하며, 특징 맵 추출기에서는 시선 방향 추정을 위한 눈 이미지의 특징 맵을 생성한다. 그리고 시선 방향 추정기에서는 각 출력 결과를 조합하고 이를 통해 최종 시선 방향 벡터를 추정한다. 제안하는 네트워크는 UnityEyes[2] 데이터셋을 통해 생성된 가상의 합성 눈 이미지와 랜드마크 좌표 데이터를 이용하여 학습하였으며, 성능 평가는 실제 사람의 눈 이미지로 구성된 MPIIGaze[3] 데이터 셋을 이용하였다. 실험을 통해 시선 추정 오차는 0.0396 MSE(Mean Square Error)의 성능을 보였으며, 네트워크의 추정 속도는 42 FPS(Frame Per Second)를 나타내었다.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.15 no.7
• /
• pp.4508-4515
• /
• 2014

Real-time body detection has been researched actively. On the other hand, the detection rate of color distorted images is low because most existing detection methods use static skin color model. Therefore, this paper proposes a new method for detecting the skin color region using a gradient map and saturation features. The basic procedure of the proposed method sequentially consists of creating a gradient map, extracting a gradient feature of skin regions, noise removal using the saturation features of skin, creating a cluster for extraction regions, detecting skin regions using cluster information, and verifying the results. This method uses features other than the color to strengthen skin detection not affected by light, race, age, individual features, etc. The results of the detection rate showed that the proposed method is 10% or more higher than the traditional methods.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.18 no.7
• /
• pp.350-356
• /
• 2017

With increasing attention to real-time body detection, active research is being conducted on human body detection based on skin color. Despite this, most existing skin detection methods utilize static skin color models and have detection rates in images, in which colors are distorted. This study proposed a method of detecting the skin region using a fuzzy classification of the gradient map, saturation, and Cb and Cr in the YCbCr space. The proposed method, first, creates a gradient map, followed by a saturation map, CbCR map, fuzzy classification, and skin region binarization in that order. The focus of this method is to rigorously detect human skin regardless of the lighting, race, age, and individual differences, using features other than color. On the other hand,the borders between these features and non-skin regions are unclear. To solve this problem, the membership functions were defined by analyzing the relationship between the gradient, saturation, and color features and generate 108 fuzzy rules. The detection accuracy of the proposed method was 86.35%, which is 2~5% better than the conventional method.