• 대한임베디드공학회논문지
• /
• 제19권1호
• /
• pp.9-17
• /
• 2024

It is widely recognized that for 3D human pose estimation (HPE), dataset acquisition is expensive and the effectiveness of augmentation techniques of conventional visual recognition tasks is limited. We address these difficulties by presenting a simple but effective method that augments input images in terms of viewpoints when training a 3D human pose estimation (HPE) model. Our intuition is that meaningful variants of the input images for HPE could be obtained by viewing a human instance in the images from an arbitrary viewpoint different from that in the original images. The core idea is to synthesize new images that have self-occlusion and thus are difficult to predict at different viewpoints even with the same pose of the original example. We incorporate this idea into the training procedure of the 3D HPE model as an augmentation stage of the input samples. We show that a strategy for augmenting the synthesized example should be carefully designed in terms of the frequency of performing the augmentation and the selection of viewpoints for synthesizing the samples. To this end, we propose a new metric to measure the prediction difficulty of input images for 3D HPE in terms of the distance between corresponding keypoints on both sides of a human body. Extensive exploration of the space of augmentation probability choices and example selection according to the proposed distance metric leads to a performance gain of up to 6.2% on Human3.6M, the well-known pose estimation dataset.

• Journal of the Korean Data and Information Science Society
• /
• 제21권2호
• /
• pp.345-354
• /
• 2010

The Gamma Distribution is widely used in Engineering and Industrial applications. Estimation of parameters is revisited in the two-parameter Gamma distribution. The parameters are estimated by minimizing the likelihood ratios. A comparative study between the method of moments, the maximum likelihood method, the method of product spacings, and minimization of three different likelihood ratios is performed using simulation. For the scale parameter, the maximum likelihood estimate performs better and for the shape parameter, the product spacings estimate performs better. Among the three likelihood ratio statistics considered, the Anderson-Darling statistic has inferior performance compared to the Cramer-von-Misses statistic and the Kolmogorov-Smirnov statistic.

• 한국공작기계학회논문집
• /
• 제13권1호
• /
• pp.22-27
• /
• 2004

A new method that estimates the sensing ability of HH smart sensor is proposed. The new signal processing method have been developed that can distinguish among different materials relatively. The HH smart sensor was developed far recognition of materials. The HH smart sensor was made for experiment. Then, it was estimated the ability to recognize objects according to acceleration value. The sensing ability of HH smart sensor has been estimated with the $R_{SAI}$ method. Experiments and analysis were executed to estimate the ability to recognize objects according to acceleration value changing. Dynamic characteristics of HH smart sensor were evaluated relatively through a new $R_{SAI}$ method that uses the power spectrum density. Applications of this method are for finding abnormal conditions of objects (auto-manufacturing), feeling of objects (medical product), robotics, safety diagnosis of structure, etc.

• 대한기계학회:학술대회논문집
• /
• 대한기계학회 2001년도 추계학술대회논문집A
• /
• pp.527-532
• /
• 2001

In this paper, we will propose the new method that estimates the sensing ability of HH smart sensor. We have developed a new signal processing method that can distinguish among different materials relatively. The HH smart sensor was developed for recognition of materials. We made the HH smart sensor in our experiment. Then, we estimated the ability to recognize objects according to acceleration value. We estimated the sensing ability of HH smart sensor with the $R_{SAI}$ method. Experiments and analysis were executed to estimate the ability to recognize objects according to acceleration value changing. Dynamic characteristics of HH smart sensor were evaluated relatively through a new $R_{SAI}$ method that uses the power spectrum density. Applications of this method are for finding abnormal conditions of objects (auto-manufacturing), feeling of objects (medical product), robotics, safety diagnosis of structure, etc.

• 한국자동차공학회논문집
• /
• 제17권1호
• /
• pp.130-136
• /
• 2009

A lumped parameter model of Li-ion battery in hybrid electric vehicle(HEV) is constructed and system parameters are identified by using recursive least square estimation for different C-rates, SOCs and temperatures. The system characteristics of pole and zero in frequency domain are analyzed with the parameters obtained from different conditions. The parameterized model of Li-ion battery indicates highly dependant of temperatures. The system pole and internal resistance changes 6.6 and 18 times at $-20^{\circ}C$, comparing with those at $25^{\circ}C$, respectively. These results will be utilized on constructing model-based state observer or an on-line identification and an adaptation of the model parameters in battery management systems for hybrid electric vehicle applications.

• Communications for Statistical Applications and Methods
• /
• 제26권5호
• /
• pp.445-461
• /
• 2019

It is possible that data are not always fitted with sufficient precision by the existing distributions; therefore this article presents a methodology that enables the use of families of asymmetric distributions as alternative probabilistic models for survival analysis, with censorship on the right, different from those usually studied (the Exponential, Gamma, Weibull, and Lognormal distributions). We use a more flexible parametric model in terms of density behavior, assuming that data can be fit by a distribution of stable distribution families considered unconventional in the analyses of survival data that are appropriate when extreme values occur, with small probabilities that should not be ignored. In the methodology, the determination of the analytical expression of the risk function h(t) of the $L{\acute{e}}vy$ distribution is included, as it is not usually reported in the literature. A simulation was conducted to evaluate the performance of the candidate distribution when modeling survival times, including the estimation of parameters via the maximum likelihood method, survival function ${\hat{S}}$(t) and Kaplan-Meier estimator. The obtained estimates did not exhibit significant changes for different sample sizes and censorship fractions in the sample. To illustrate the usefulness of the proposed methodology, an application with real data, regarding the survival times of patients with colon cancer, was considered.

• 한국콘텐츠학회논문지
• /
• 제12권4호
• /
• pp.31-39
• /
• 2012

분산 비디오 압축 기술은 초경량 비디오 압축 기술로써 많은 주목을 받고 있으며, 대표적인 기법은 피드백 채널을 이용하여 우수한 부호화 성능을 유지한다. 그러나 이로 인해 복호화기의 복잡도를 증대시키고 매우 많은 반복적인 연산에 의한 큰 복호화 지연을 요구하기 때문에 실시간 구현에 제한이 되고 있으며, 이를 개선하기 위한 연구가 필요하다. 이를 위해, 본 논문에서는 화소 영역 위너-지브 비디오 복호화 기법에서 각 비트 플레인 내에 위치한 LDPCA 프레임간의 시간적 상관성, 공간적 상관성 그리고 시공간적 상관성 등을 고려한 패리티 비트 요구량에 대한 예측 방법을 제시하고 고속 분산 비디오 복호화기법에 적용하여 성능을 비교한다. 모의실험을 통해, 움직임이 큰 영상과 움직임이 적은 영상에 대해 각각 시공간적 상관성과 시간적 상관성을 이용한 방식이 우수한 특성을 보이며, 이는 분산 비디오 부호화 기법의 다양한 응용 환경에 따른 효과적인 패리티 요구량 예측기법을 찾는데 효과적으로 사용될 수 있을 것이다.

• 한국정보통신학회논문지
• /
• 제15권8호
• /
• pp.1659-1664
• /
• 2011

Fuzzy C-Means (FCM)는 군집화를 위해 널리 사용되는 알고리듬 중 하나로 다양한 응용 분야에서 성공적으로 사용되어 왔다. 하지만 FCM은 여러 가지 단점을 가지고 있으며 초기 원형 설정이 그 중 하나이다. FCM은 국부 최적해에 수렴하므로 초기 원형 설정에 따라 군집화의 결과가 달라진다. 따라서 초기 원형의 설정은 군집화 결과 향상을 위해 중요하다. 이 논문에서는 이러한 FCM의 초기 원형 설정 문제를 해결하는 방안으로 커널 밀도 추정을 활용하는 방법을 제안한다. 커널 밀도 추정은 비모수적 분포들에도 사용할 수 있어 국부적인 데이터 밀도 추정에 유용하다. 제안한 방법에서는 커널 밀도 추정을 수행한 후 밀도가 높은 지역에 클러스터의 초기 원형을 설정하고 원형이 설정된 영역의 밀도를 감소시키는 과정을 반복함으로써 효율적으로 초기 원형을 선택할 수 있다. 제안된 방법이 일반적으로 사용되는 무작위 초기화 방법에 비해 효율적이라는 사실은 실험 결과를 통해 확인할 수 있다.

• 한국소음진동공학회:학술대회논문집
• /
• 한국소음진동공학회 2001년도 춘계학술대회논문집
• /
• pp.318-322
• /
• 2001

This paper deals with sensing ability of smart sensor that has a sensing ability to distinguish materials according to surface types of smart sensor. We have developed a new signal processing method that can distinguish among different materials. The smart sensor was developed for recognition of materials. We made two types of smart sensors in our experiment. Then, we estimated the ability to recognize objects according to smart sensor type. We estimated the sensing ability of smart sensor with the $R_{SAI}$ method. Experiments and analysis were executed to estimate the ability to recognize objects according to surface types of smart sensor. Sensing ability of smart sensors was evaluated relatively through a new $R_{SAI}$ method. Applications of smart sensors are for finding abnormal conditions of objects (auto-manufacturing), feeling of objects (medical product), robotics, safety diagnosis of structure, etc.etc.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제9권8호
• /
• pp.3136-3150
• /
• 2015

A vision-based 3D tracking of articulated human hand is one of the major issues in the applications of human computer interactions and understanding the control of robot hand. This paper presents an improved approach for tracking and recovering the 3D position and orientation of a human hand using the Kinect sensor. The basic idea of the proposed method is to solve an optimization problem that minimizes the discrepancy in 3D shape between an actual hand observed by Kinect and a hypothesized 3D hand model. Since each of the 3D hand pose has 23 degrees of freedom, the hand articulation tracking needs computational excessive burden in minimizing the 3D shape discrepancy between an observed hand and a 3D hand model. For this, we first created a 3D hand model which represents the hand with 17 different parts. Secondly, Random Forest classifier was trained on the synthetic depth images generated by animating the developed 3D hand model, which was then used for Haar-like feature-based classification rather than performing per-pixel classification. Classification results were used for estimating the joint positions for the hand skeleton. Through the experiment, we were able to prove that the proposed method showed improvement rates in hand part recognition and a performance of 20-30 fps. The results confirmed its practical use in classifying hand area and successfully tracked and recovered the 3D hand pose in a real time fashion.