• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.485-490
• /
• 2022

데이터 증강기법은 추가적인 데이터 구축 혹은 수집 행위 없이 원본 데이터셋의 양과 다양성을 증가시키는 방법이다. 데이터 증강기법은 규칙 기반부터 모델 기반 방법으로 발전하였으며, 최근에는 Masked Language Modeling (MLM)을 응용한 모델 기반 데이터 증강 연구가 활발히 진행되고 있다. 그러나 기존의 MLM 기반 데이터 증강 방법은 임의 대체 방식을 사용하여 문장 내 의미 변화 가능성이 큰 주요 토큰을 고려하지 않았으며 증강에 따른 레이블 교정방법이 제시되지 않았다는 한계점이 존재한다. 이러한 문제를 완화하기 위하여, 본 논문은 레이블을 고려할 수 있는 Re-labeling module이 추가된 MLM 기반 한국어 데이터 증강 방법론을 제안한다. 제안하는 방법론을 KLUE-STS 및 KLUE-NLI 평가셋을 활용하여 검증한 결과, 기존 MLM 방법론 대비 약 89% 적은 데이터 양으로도 baseline 성능을 1.22% 향상시킬 수 있었다. 또한 Gate Function 적용 여부 실험으로 제안 방법 Re-labeling module의 구조적 타당성을 검증하였다.

• Journal of Electrical Engineering and Technology
• /
• v.11 no.6
• /
• pp.1857-1862
• /
• 2016

Human activity recognition using depth information is an emerging and challenging technology in computer vision due to its considerable attention by many practical applications such as smart home/office system, personal health care and 3D video games. This paper presents a novel framework of 3D human body detection, tracking and recognition from depth video sequences using spatiotemporal features and modified HMM. To detect human silhouette, raw depth data is examined to extract human silhouette by considering spatial continuity and constraints of human motion information. While, frame differentiation is used to track human movements. Features extraction mechanism consists of spatial depth shape features and temporal joints features are used to improve classification performance. Both of these features are fused together to recognize different activities using the modified hidden Markov model (M-HMM). The proposed approach is evaluated on two challenging depth video datasets. Moreover, our system has significant abilities to handle subject's body parts rotation and body parts missing which provide major contributions in human activity recognition.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.25 no.1
• /
• pp.62-69
• /
• 2014

In this paper, WBAN(Wireless Body Area Network) On-body system using the surface-oriented antenna about the impact of human internal organs were analyzed through experiments. The received signal strength is measured for effect of human using the human model and the phantom of torso. Experiments are performed in anechoic chamber without moving and measured by Vector Network Analyzer. This paper confirms the effect of human body by comparing the human model and the phantom of torso. And also know the human internal organs effect on the antennas loss of received signal strength by measured data.

• Journal of Korea Multimedia Society
• /
• v.11 no.6
• /
• pp.860-868
• /
• 2008

Human detection techniques in outdoor scenes have been studied for a long time to watch suspicious movements or to keep someone from danger. However there are few methods of human detection in overhead or near-field view scenes, while lots of human detection methods in far-field view scenes have been developed. In this paper, a set of five features useful for human detection in overhead view scenes and another set of four useful features in near-field view scenes are suggested. Eight feature-candidates are first extracted by analyzing geometrically varying characteristics of moving objects in samples of video sequences. Then highly contributed features for each view scene to classifying human from other moving objects are selected among them by using a neural network learning technique. Through experiments with hundreds of moving objects, we found that each set of features is very useful for human detection and classification accuracy for overhead view and near-field view scenes was over 90%. The suggested sets of features can be used effectively in a PTZ camera based surveillance system where both the overhead and near-field view scenes appear.

• 제어로봇시스템학회:학술대회논문집
• /
• 2004.08a
• /
• pp.800-803
• /
• 2004

This paper presents a wavelet-based digital watermarking technique for still images. The digital watermarking considering human visual system (HVS) to increase the robustness and perceptual invisibility of digital watermark. The watermarking embedding is modified discrete wavelet transform (DWT) coefficients of the subbands of the images. The human visual system is number of factors that effect the noise sensitivity of human eyes that is considered to increase the robustness and perceptual invisibility of digital watermark. The watermark detection is blind watermark ( original image is not required ). Experimental results successful against attacks by image processing such as add noise, cropping, filtering, JPEG and JPEG2000 compression.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 2004.10a
• /
• pp.1053-1056
• /
• 2004

Research in Human Computer Interface (HCI) is towards development of an application environment able to deal with interactions of both human and computers that can be more intuitive and efficient. This can be achieved by bridging the gap between the synthetic virtual environment and the natural physical environment. Thus a project called Tangible Space Initiative (TSI) has been launched by KIST. TSI is subdivided into Tangible Interface (TI) which controls 3D cyber space with user's perspective, Responsive Cyber Space (RCS) which creates and controls the virtual environment and Tangible Agent (TA) which senses and acts upon the physical interface environment on behalf of any components of TSI or the user. This paper is a brief introduction to a new generation of Human Computer Interface that bring user to a new era of interaction with computers in the future.

• 제어로봇시스템학회:학술대회논문집
• /
• 2005.06a
• /
• pp.1025-1030
• /
• 2005

This paper deals with the emotional model of the software-robot. The software-robot requires several capabilities such as sensing, perceiving, acting, communicating, and surviving. and so on. There are already many studies about the emotional model like KISMET and AIBO. The new emotional model using the modified friendship scheme is proposed in this paper. Quite often, the available emotional models have time invariant human respond architectures. Conventional emotional models make the sociable robot get around with humans, and obey human commands during robot operation. This behavior makes the robot very different from real pets. Similar to real pets, the proposed emotional model with the modified friendship capability has time varying property depending on interaction between human and robot.

• Journal of the Korea Society of Computer and Information
• /
• v.10 no.5 s.37
• /
• pp.171-177
• /
• 2005

This paper describes the method of calculating coordinate using Forward Kinematics and expresses the recursive equation as the numerical formula using a homogeneous coordinate for creating workspace. This paper proposes an algorithm for the workspace of human model using the recursive equation and the joint limit angle of human model, and describes the results of workspace of the human model as computer graphics.

• Journal of Korea Multimedia Society
• /
• v.22 no.4
• /
• pp.432-442
• /
• 2019

Most current human action recognition methods based on deep learning methods. It is required, however, a very high computational cost. In this paper, we propose an action change detection method to reduce repetitive human action recognition tasks. In reality, simple actions are often repeated and it is time consuming process to apply high cost action recognition methods on repeated actions. The proposed method decides whether action has changed. The action recognition is executed only when it has detected action change. The action change detection process is as follows. First, extract the number of non-zero pixel from motion history image and generate one-dimensional time-series data. Second, detecting action change by comparison of difference between current time trend and local extremum of time-series data and threshold. Experiments on the proposed method achieved 89% balanced accuracy on action change data and 61% reduced action recognition repetition.

• International Journal of Computer Science & Network Security
• /
• v.21 no.2
• /
• pp.221-228
• /
• 2021

Recent improvements on the quality, fidelity and availability of biometric data have led to effective human physical activity detection (HPAD) in real time which adds significant value to applications such as human behavior identification, healthcare monitoring, and user authentication. Current approaches usually use machine-learning techniques for human physical activity recognition based on the data collected from wearable accelerometer sensor from a single wearable smart device on the user. However, collecting data from a single wearable smart device may not provide the complete user activity data as it is usually attached to only single part of the user's body. In addition, in case of the absence of the single sensor, then no data can be collected. Hence, in this paper, a continuous HPAD will be presented to effectively perform user activity detection with mobile service infrastructure using multiple wearable smart devices, namely smartphone and smartwatch placed in various locations on user's body for more accurate HPAD. A case study on a comprehensive dataset of classified human physical activities with our HAPD approach shows substantial improvement in HPAD accuracy.