• 한국지능시스템학회논문지
• /
• 제25권1호
• /
• pp.29-34
• /
• 2015

Brain-computer interface 기술은 일상에서 편안한 생활을 위해 다방면으로 연구가 진행 중이다. 본 연구는 3가지 동작의 뇌파특성을 분석하여 의수와 같은 외부기기의 세밀한 동작 제어를 목적으로 한다. 피험자들은 악력기를 쥘 때 (Grip), 손가락만을 움직일 때 (Move), 아무런 동작을 취하지 않을 때 (Relax)의 3가지 동작을 수행하였고, 뇌파를 측정하여 power spectrum analysis와 multi-common spatial pattern 알고리즘으로 특징추출을 수행하였으며, 분류알고리즘인 SVM(support vector machine)으로 뇌파의 특징데이터들을 분류하였다. 실험결과 3개의 다른 동작을 분류한 결과, 실험에 참여한 3명의 피험자 중 2명에게서 Grip 클래스의 분류율이 가장 높은 분류율을 보였다. 본 연구의 결과는 뇌파를 이용하여 의수가 필요한 환자들에게 유용할 것으로 기대한다.

• 대한전자공학회논문지SP
• /
• 제43권4호
• /
• pp.74-86
• /
• 2006

본 논문에서는 계산 효율적이고 노이즈에 강건한 비디오 객체 분할 알고리즘을 제안한다. 움직임 분할과 색 분할을 효율적으로 결합한 시공간 분할 방법의 구현을 위해 SOM 기반의 계층적 군집 방법을 도입하여 특징 벡터들의 군집 관점에서 분할 과정을 해석함으로써 기존의 객체 분할 방법에서 정확한 분할 결과를 얻기 위해서 요구되어지는 많은 연산량과 노이즈에 의한 시스템의 성능 저하 문제를 최소화한다. 움직임 분할 과정에서는 움직임 추정 에러에 의한 영향을 최소화하기 위해서 MRF 기반의 MAP 추정 방법을 이용하여 계산한 움직임 벡터의 신뢰도를 이용한다. 또한 움직임 분할의 성능 향상을 위해서 움직임 신뢰도 히스토그램을 이용한 노이즈 제거 과정을 거칠 뿐만 아니라 자동으로 장면 내에 존재하는 객체의 수를 구하기 위해서 군집 유효성 지표를 이용한다. 객체 추적의 성능 향상을 위해 교차 투영 기법을 이용하며, 분할 결과의 시간적 일관성 유지를 위해 동적 메모리를 이용한다. 다양한 특성을 가지는 비디오 시퀀스들을 이용한 실험을 통해 제안하는 방법이 계산 효율적이고 노이즈에 강건하게 비디오 객체 분할을 수행함은 물론 기존의 구현 방법에 비해 정확한 분할 결과를 얻을 수 있음을 확인하였다.

• IEIE Transactions on Smart Processing and Computing
• /
• 제2권6호
• /
• pp.345-349
• /
• 2013

This paper proposes a simplified generalized residual prediction (GRP) that reduces the computational complexity of spatial scalability in scalable high efficiency video coding (SHVC). GRP is a coding tool to improve the inter prediction by adding a residual signal to the inter predictor. The residual signal was created by carrying out motion compensation (MC) of both the enhancement layer (EL) and up-sampled reference layer (RL) with the motion vector (MV) of the EL. In the MC process, interpolation of the EL and the up-sampled RL are required when the MV of the EL has sub-pel accuracy. Because the up-sampled RL has few high frequency components, interpolation of the up-sampled RL does not give significantly new information. Therefore, the proposed method reduces the computational complexity of the GRP by skipping the interpolation of the up-sampled RL. The experiment on SHVC software (SHM-2.0) showed that the proposed method reduces the decoding time by 10 % compared to conventional GRP. The BD-rate loss of the proposed method was as low as 1.0% on the top of SHM-2.0.

• 한국멀티미디어학회논문지
• /
• 제14권12호
• /
• pp.1536-1543
• /
• 2011

Video stabilization is a very important step for vision based applications in the vehicular technology because the accuracy of these applications such as obstacle distance estimation, lane detection and tracking can be affected by bumpy roads and oscillation of vehicle. Conventional methods suffer from either the zooming effect which caused by a camera movement or some motion of surrounding vehicles. In order to overcome this problem, we propose a novel video stabilization method using FOE(Focus of Expansion). When a vehicle moves, optical flow diffuses from the FOE and the FOE is equal to an epipole. If a vehicle moves with vibration, the position of the epipole in the two consecutive frames is changed by oscillation of the vehicle. Therefore, we carry out video stabilization using motion vector estimated from the amount of change of the epipoles. Experiment results show that the proposed method is more efficient than conventional methods.

• International Journal of Aeronautical and Space Sciences
• /
• 제1권1호
• /
• pp.81-91
• /
• 2000

The mechanization of navigation equation is depending on the designer according to the orientation vector relating the body frame to a chosen to inertial and navigation frames for its purposes. This paper considers the appropriate Earth Fixed frame for short range vehicle and develops a mechanization and algorithm for Strapdown Inertial Navigation System(SDINS). This mechanization consists of two parts : translational mechanization and rotational mechanization{attitude determination). The accuracy, availability and performance of this SDINS mechanization are verified on the simulation and the numerical method for integration attitude propagation is compared with a well-known method in a precession motion.

• 대한기계학회논문집
• /
• 제18권7호
• /
• pp.1674-1684
• /
• 1994

An efficient iterative method is presented for the dynamic analysis of bodies moving on flexible structures. In contrast to traditional approaches, the nominal motion of the body is considered here as an unknown. The correct contact forces between the bodies and the flexible structures are computed by an iterative method reducing the specially defined error vectors to zero, and thus satisfying the constraints between the bodies and the structures. Even thought only simple equations of motions and simple time integrators are adopted, the correct solutions are economically obtained and the Timoshenko paradox is completely resolved. Numerical simulations are conducted demonstrate the accuracy and reliability of the solution and to compare the results with the reference.

• 한국멀티미디어학회논문지
• /
• 제21권4호
• /
• pp.478-484
• /
• 2018

The choice of the motion features influences the result of the human action recognition method directly. Many factors often influence the single feature differently, such as appearance of the human body, environment and video camera. So the accuracy of action recognition is restricted. On the bases of studying the representation and recognition of human actions, and giving fully consideration to the advantages and disadvantages of different features, the Dense Trajectories(DT) algorithm is a very classic algorithm in the field of behavior recognition feature extraction, but there are some defects in the use of optical flow images. In this paper, we will use the improved Dense Trajectories(iDT) algorithm to optimize and extract the optical flow features in the movement of human action, then we will combined with Support Vector Machine methods to identify human behavior, and use the image in the KTH database for training and testing.

• 대한원격탐사학회:학술대회논문집
• /
• 대한원격탐사학회 2007년도 Proceedings of ISRS 2007
• /
• pp.146-148
• /
• 2007

CMDPS (COMS Meteorological Data Processing System) is the operational meteorological products extraction system for data observed from COMS (Communication, Ocean and Meteorological Satellite) meteorological imager. CMDPS baseline products consist of 16 parameters including cloud information, water vapor products, surface information, environmental products and atmospheric motion vector. Additionally, CMDPS includes the function of calibration monitoring, and validation mechanism of the baseline products. The main objective of CMDPS validation module development is near-real time monitoring for the accuracy and reliability of the whole CMDPS products. Also, its long time validation statistics are used for upgrade of CMDPS such as algorithm parameter tuning and retrieval algorithm modification. This paper introduces the preliminary design on CMDPS validation module.

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2002년도 추계학술대회 논문집
• /
• pp.1003-1006
• /
• 2002

In machining free-form curves with a machine tool equipped with parallel device, improving contouring accuracy is very important. In this paper, we present contouring control algorithm far parallel machine tool. The relation between the error in Joint space and the error in catesian space is evaluated, and we estimate contouring error vector which efficiently determines the variable gains for the cross coupled control. To show the validity of the algorithm, the contouring control is simulated for free form contour trajectory in cubic parallel machine tool model.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권4호
• /
• pp.2148-2161
• /
• 2019

In this paper, we present a real-time cattle action recognition algorithm to detect the estrus phase of cattle from a live video stream. In order to classify cattle movement, specifically, to detect the mounting action, the most observable sign of the estrus phase, a simple yet effective feature description exploiting motion history images (MHI) is designed. By learning the proposed features using the support vector machine framework, various representative cattle actions, such as mounting, walking, tail wagging, and foot stamping, can be recognized robustly in complex scenes. Thanks to low complexity of the proposed action recognition algorithm, multiple cattle in three enclosures can be monitored simultaneously using a single fisheye camera. Through extensive experiments with real video streams, we confirmed that the proposed algorithm outperforms a conventional human action recognition algorithm by 18% in terms of recognition accuracy even with much smaller dimensional feature description.