• Journal of Industrial Technology
• /
• v.29 no.A
• /
• pp.47-54
• /
• 2009

In this paper, we proposed hand detection and hand gesture recognition from USB camera video. Firstly, we extract hand region extraction using skin color information from a difference images. Background image is initially stored and extracted from the input images in order to reduce problems from complex backgrounds. After that, 16-directional chain code sequence is computed from the tracking of hand motion. These chain code sequences are compared with pre-trained models using DP matching. Our hand gesture recognition system can be used to control PowerPoint slides or applied to multimedia education systems. We got 92% hand region extraction accuracy and 82.5% gesture recognition accuracy, respectively.

• Proceedings of the Korean Institute of Electrical and Electronic Material Engineers Conference
• /
• 2003.05c
• /
• pp.194-199
• /
• 2003

Depth error correction effect for maladjusted stereo cameras with calibrated pixel distance parameter is presented. The proposed neural network technique is the real time computation method based theory of inter-node diffusion for searching the safety distances from the sudden appearance-objects during the work driving. The main steps of the distance computation using the theory of stereo vision like the eyes of man is following steps. One is the processing for finding the corresponding points of stereo images and the other is the interpolation processing of full image data from nonlinear image data of objects. All of them request much memory space and time. Therefore the most reliable neural-network algorithm is derived for real-time matching of objects, which is composed of a dynamic programming algorithm based on sequence matching techniques.

• Proceedings of the KIEE Conference
• /
• 1996.07b
• /
• pp.1350-1352
• /
• 1996

The purpose of image sequence coding is to reduce the spatio-temporal redundancies. For the spatial redundancies, we can use the transform coding such as DCT. In this paper, the optical flow method is applied to solve the problem of temporal redundancies. There are several estimation methods like block matching method and pel-recursive method. Block matching method is easy for a hardware implementation because of the computational simplicity. So, it is now used as the estimation method in MPEG-l, MPEG-2, and H.261. We compared the merits and demerits of the optical flow method and the block matching method in this paper.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2012.04a
• /
• pp.224-227
• /
• 2012

시계열 데이터 검색은 금융, 생명정보 등 많은 분야에서 요구되는 주요한 기술로서 연구되어 왔다. 특히 기존에 제안된 스케일링(Scaling)과 쉬프팅(Shifting)을 이용한 검색인 SE Match는 유사 서브시퀀스를 효과적으로 찾아내는 방법으로 알려지고 있다. 본 논문에서는 이 방법에 맵리듀스를 적용하는 MRSE-검색 (MapReduce-based Searching with Shift-Eliminated)방법을 제안한다. 본 논문이 제안하는 방법으로 분산처리를 통하여 응답시간의 개선과 대용량의 시계열 데이터에서 효율적인 검색이 가능 할 것으로 사료된다.

• Journal of KIISE:Databases
• /
• v.36 no.4
• /
• pp.327-340
• /
• 2009

To achieve the noise reduction effect in boundary image matching, we use the moving average transform of time-series matching. Our motivation is based on an intuition that using the moving average transform we may exploit the noise reduction effect in boundary image matching as in time-series matching. To confirm this simple intuition, we first propose $\kappa$-order image matching, which applies the moving average transform to boundary image matching. A boundary image can be represented as a sequence in the time-series domain, and our $\kappa$-order image matching identifies similar images in this time-series domain by comparing the $\kappa$-moving average transformed sequences. Next, we propose an index-based matching method that efficiently performs $\kappa$-order image matching on a large volume of image databases, and formally prove the correctness of the index-based method. Moreover, we formally analyze the relationship between an order $\kappa$ and its matching result, and present a systematic way of controlling the noise reduction effect by changing the order $\kappa$. Experimental results show that our $\kappa$-order image matching exploits the noise reduction effect, and our index-based matching method outperforms the sequential scan by one or two orders of magnitude.

• Proceedings of the Korean Information Science Society Conference
• /
• 2007.10a
• /
• pp.7-8
• /
• 2007

• Annual Conference on Human and Language Technology
• /
• 2017.10a
• /
• pp.35-40
• /
• 2017

기계 독해(Machine reading comprehension)는 주어진 문맥을 이해하고, 질문에 적합한 답을 문맥 내에서 찾는 문제이다. Simple Recurrent Unit (SRU)은 Gated Recurrent Unit (GRU)등과 같이 neural gate를 이용하여 Recurrent Neural Network (RNN)에서 발생하는 vanishing gradient problem을 해결하고, gate 입력에서 이전 hidden state를 제거하여 GRU보다 속도를 향상시킨 모델이며, Self-matching Network는 R-Net 모델에서 사용된 것으로, 자기 자신의 RNN sequence에 대하여 어텐션 가중치 (attention weight)를 계산하여 비슷한 의미 문맥 정보를 볼 수 있기 때문에 상호참조해결과 유사한 효과를 볼 수 있다. 본 논문에서는 한국어 기계 독해 데이터 셋을 구축하고, 여러 층의 SRU를 이용한 Encoder에 Self-matching layer를 추가한 $S^2$-Net 모델을 제안한다. 실험 결과, 본 논문에서 제안한 $S^2$-Net 모델이 한국어 기계 독해 데이터 셋에서 EM 65.84%, F1 78.98%의 성능을 보였다.

• 한국어정보학회:학술대회논문집
• /
• 2017.10a
• /
• pp.35-40
• /
• 2017

기계 독해(Machine reading comprehension)는 주어진 문맥을 이해하고, 질문에 적합한 답을 문맥 내에서 찾는 문제이다. Simple Recurrent Unit (SRU)은 Gated Recurrent Unit (GRU)등과 같이 neural gate를 이용하여 Recurrent Neural Network (RNN)에서 발생하는 vanishing gradient problem을 해결하고, gate 입력에서 이전 hidden state를 제거하여 GRU보다 속도를 향상시킨 모델이며, Self-matching Network는 R-Net 모델에서 사용된 것으로, 자기 자신의 RNN sequence에 대하여 어텐션 가중치 (attention weight)를 계산하여 비슷한 의미 문맥 정보를 볼 수 있기 때문에 상호참조해결과 유사한 효과를 볼 수 있다. 본 논문에서는 한국어 기계 독해 데이터 셋을 구축하고, 여러 층의 SRU를 이용한 Encoder에 Self-matching layer를 추가한 $S^2$-Net 모델을 제안한다. 실험 결과, 본 논문에서 제안한 $S^2$-Net 모델이 한국어 기계 독해 데이터 셋에서 EM 65.84%, F1 78.98%의 성능을 보였다.

• Journal of KIISE:Computing Practices and Letters
• /
• v.14 no.6
• /
• pp.559-566
• /
• 2008

In this paper, we propose a range subsequence matching under dynamic time warping (DTW) distance. We exploit Dual Match, which divides data sequences into disjoint windows and the query sequence into sliding windows. However, Dual Match is known to work under Euclidean distance. We argue that Euclidean distance is a fragile distance, and thus, DTW should be supported by Dual Match. For this purpose, we derive a new important theorem showing the correctness of our approach and provide a detailed algorithm using the theorem. Extensive experimental results show that our range subsequence matching performs much better than the sequential scan algorithm.

• Journal of Institute of Control, Robotics and Systems
• /
• v.3 no.5
• /
• pp.520-525
• /
• 1997

This paper presents a fault detection and diagnosis methodology based on weighted symptom model and pattern matching between the coming fault propagation trend and the simulated one. In the first step, backward chaining is used to find the possible cause candidates for the faults. The weighted symptom model is used to generate those candidates. The weight is determined from dynamic simulation. Using WSM, the methodology can generate the cause candidates and rank them according to the probability. Second, the fault propagation trends identified from the partial or complete sequence of measurements are compared with the standard fault propagation trends stored a priori. A pattern matching algorithm based on a number of triangular episodes is used to effectively match those trends. The standard trends have been generated using dynamic simulation and stored a priori. The proposed methodology has been illustrated using two case studies, and the results showed satisfactory diagnostic resolution.