• Smart Media Journal
• /
• v.12 no.7
• /
• pp.59-67
• /
• 2023

Recently, research on behavior analysis tracking human posture and movement has been actively conducted. In particular, OpenPose, an open-source software developed by CMU in 2017, is a representative method for estimating human appearance and behavior. OpenPose can detect and estimate various body parts of a person, such as height, face, and hands in real-time, making it applicable to various fields such as smart healthcare, exercise training, security systems, and medical fields. In this paper, we propose a method for classifying four exercise movements - Squat, Walk, Wave, and Fall-down - which are most commonly performed by users in the gym, using OpenPose-based deep learning models, DNN and CNN. The training data is collected by capturing the user's movements through recorded videos and real-time camera captures. The collected dataset undergoes preprocessing using OpenPose. The preprocessed dataset is then used to train the proposed DNN and CNN models for exercise movement classification. The performance errors of the proposed models are evaluated using MSE, RMSE, and MAE. The performance evaluation results showed that the proposed DNN model outperformed the proposed CNN model.

• Journal of the Institute of Electronics Engineers of Korea SC
• /
• v.45 no.6
• /
• pp.123-131
• /
• 2008

In Oriental medicine, the status of a tongue is the important indicator to diagnose the condition of one's health like the physiological and the clinicopathological changes of internal organs in a body. A tongue diagnosis is not only convenient but also non-invasive, and therefore widely used in Oriental medicine. However, the tongue diagnosis is affected by examination circumstances like a light source, patient's posture, and doctor's condition a lot. To develop an automatic tongue diagnosis system for an objective and standardized diagnosis, segmenting a tongue region from a facial image captured and classifying tongue coating are inevitable but difficult since the colors of a tongue, lips, and skin in a mouth are similar. The proposed method includes preprocessing, over-segmenting, detecting the edge with a local minimum over a shading area from the structure of a tongue, correcting local minima or detecting the edge with the greatest color difference, selecting one edge to correspond to a tongue shape, and smoothing edges, where preprocessing consists of down-sampling to reduce computation time, histogram equalization, and edge enhancement, which produces the region of a segmented tongue. Finally, the systematic procedure separated only a tongue region from a face image with a tongue, which was obtained from a digital tongue diagnosis system. Oriental medical doctors' evaluation for the results illustrated that the segmented region excluding a non-tongue region provides important information for the accurate diagnosis. The proposed method can be used for an objective and standardized diagnosis and for an u-Healthcare system.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.49 no.3
• /
• pp.30-39
• /
• 2012

In this paper, a safe driving assistance system is proposed by detecting the status of driver's doze based on face and eye detection. By the level of the fatigue, safe driving system alarms or set the seatbelt on vibration. To reduce the effect of backward light and too strong solar light which cause a decrease of face and eye detection rate and false fatigue detection, post processing techniques like image equalization are used. Haar transform and PCA are used for face detection. By using the statistic of the face and eye structural ratio of normal Koreans, we can reduce the eye candidate area in the face, which results in reduction of the computational load. We also propose a new eye status detection algorithm based on Hough transform and eye width-height ratio, which are used to detect eye's blinking status which decides doze level by measuring the blinking period. The system alarms and operates seatbelt on vibration through controller area network(CAN) when the driver's doze level is detected. In this paper, four algorithms are implemented and proposed algorithm is made based on the probability model and we achieves 84.88% of correct detection rate through indoor and in-car environment experiments. And also we achieves 69.81% of detection rate which is better result than that of other algorithms using IR camera.

• The Journal of the Korea Contents Association
• /
• v.9 no.5
• /
• pp.1-12
• /
• 2009

The correct detection of skin color is an important preliminary process in fields of face detection and human motion analysis. It is generally performed by three steps: transforming the pixel color to a non-RGB color space, dropping the illuminance component of skin color, and classifying the pixels by the skin color distribution model. Skin detection depends on by various factors such as color space, presence of the illumination, skin modeling method. In this paper we propose a 3d skin color model that can segment pixels with several ethnic skin color from images with various illumination condition and complicated backgrounds. This proposed skin color model are formed with each components(Y, Cb, Cr) which transform pixel color to YCbCr color space. In order to segment the skin color of several ethnic groups together, we first create the skin color model of each ethnic group, and then merge the skin color model using its skin color probability. Further, proposed model makes several steps of skin color areas that can help to classify proper skin color areas using small training data.

• Journal of IKEEE
• /
• v.23 no.4
• /
• pp.1250-1256
• /
• 2019

Although one can easily generate real-world 3D mesh data using a 3D printer or a depth camera, the generated data inevitably includes unnecessary noise. Therefore, mesh denoising is essential to obtain intact 3D mesh data. However, conventional mathematical denoising methods require preprocessing and often eliminate some important features of the 3D mesh. To address this problem, this paper proposes a deep learning based 3D mesh denoising method. Specifically, we propose a convolution-based autoencoder model consisting of an encoder and a decoder. The convolution operation applied to the mesh data performs denoising considering the relationship between each vertex constituting the mesh data and the surrounding vertices. When the convolution is completed, a sampling operation is performed to improve the learning speed. Experimental results show that the proposed autoencoder model produces faster and higher quality denoised data than the conventional methods.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.24 no.5
• /
• pp.504-511
• /
• 2014

This paper presents a hybrid recognition method, which is based on the nonlinear histogram equalization and the multidimensional intensity variation of an images. The nonlinear histogram equalization based on a adaptively modified function is applied to improve the quality by adjusting the brightness of the image. The multidimensional intensity variation by considering the a extent of 4-step changes in brightness between the adjacent pixels is also applied to reflect accurately the attributes of image. The statistical correlation that is measured by the normalized cross-correlation(NCC) coefficient, is applied to comprehensively measure the similarity between the images. The NCC is considered by the intensity variation of each 2-direction(x-axis and y-axis) image. The proposed method has been applied to the problem for recognizing the 50-face images of 40*40 pixels. The experimental results show that the proposed method has a superior recognition performances to the method without performing the histogram equalization, or the linear histogram equalization, respectively.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.28 no.1
• /
• pp.85-93
• /
• 2018

Biometric user authentications, day after day, propagate more to human life instead of traditional systems which use passwords and ID cards. However, most of these systems have many problems for given biometric information such noisy data, low-quality data, a limitation of recognition rate, and so on. To deal with these problems, I used Dual QML which is non-linear classification for classifying correctly the real-world data and then proposed preprocessing method for increasing recognition rate and performance by segmenting a specific region on an image. The previous published Dual QML used face, palmprint, ear for the experiment. In this paper, I used iris for experiment and then proved excellence of Dual QML at iris recognition. Finally I demonstrated these results (e.g. increasing recognition rate and performance, suitability for iris recognition) through experiments.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.8 no.1
• /
• pp.50-59
• /
• 2007

In this paper, an anchor shot detection method, which is a basic technology for managing news videos for index and retrieval purposes is proposed. To do that, two most popular news program such as 'KBS 9 Hour News' and 'MBC 9 Hour News' are analyzed and 4-step rule based detection method is proposed First, in the preprocessing, video shot boundaries are detected and the 1st frame of each shot is extracted as a key frame. Then, the detected shot is declared as an anchor shot, if all the following 4 conditions are satisfied. 1) There is an anchor face in the key frame of a shot. 2) Spatial distribution of edges in the key frame is adequate. 3) Background color information of the key frame is similar to the color information of an anchor model. 4) Motion rate in the shot is low. In order to show the validity of the proposed method, three 'KBS 9 Hour News' and three 'MBC 9 Hour News', which have total running time of 108 in minute and are broadcasted at different days, are used for experiments. Average detection rates showed 0.97 in precision, 1.0 in recall, and 0.98 in F-measure.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.11a
• /
• pp.565-567
• /
• 2022

질병을 앓고 있는 환자는 상태에 따라 병실, 주거지, 요양원 등 조호환경 내 생활 시 의료 인력의 지속적인 추적 및 관찰을 통해 신체에 이상이 생긴 경우 이를 감지하고, 신속하게 조치할 수 있도록 해야 한다. 의료 인력이 직접 환자를 확인하는 방법은 의료 인력의 반복적인 노동이 요구되며 실시간으로 환자를 확인해야 한다는 특성상 의료 인력이 상주해야 하기에 이는 곧, 의료 인력의 부족과 낭비로 이어진다. 해당 문제 해결을 위해 의료 인력을 대신하여 조호환경 내 환자의 상태를 실시간으로 모니터링할 수 있는 딥러닝 모델들이 연구되고 있다. 딥러닝 모델은 데이터의 수가 많을수록 강인한 모델을 설계할 수 있으며, 데이터셋의 배경, 객체의 특징 분포 등 다양한 조건에 영향을 받기 때문에 학습에 필요한 도메인을 가지는 많은 양의 전처리된 데이터를 수집해야 한다. 따라서, 조호환경 내 환자에 대한 데이터셋이 필요하지만, 공개된 데이터셋의 경우 양이 매우 적으며 이를 반전, 회전기법 등을이용할 경우 데이터의 수를 늘릴 수 있지만, 같은 분포의 특징을 가지는 데이터가 생성되기에 데이터 증강 기법을 단순하게 적용하면 딥러닝 모델의 과적합을 야기한다. 또한, 조호환경 내 이미지 데이터셋은 얼굴 노출과 같은 개인정보가 포함 될 수 있으며 이를 보호하기 위해 정보들을 비식별화 해야 한다는 문제점이 있다. 따라서 본 논문에서는 조호환경에서 수집된 데이터 증강을 위한 Style-Generative Adversarial Networks 기법을 적용하여 조호환경 데이터셋 수집에 효과적인 증강 기법을 제안한다.

• Journal of Broadcast Engineering
• /
• v.17 no.3
• /
• pp.519-528
• /
• 2012

Voice activity detection (VAD) is generally conducted by extracting features from the acoustic signal and a decision rule. The performance of such VAD algorithms driven by the input acoustic signal highly depends on the acoustic noise. When video signals are available as well, the performance of VAD can be enhanced by using the visual information which is not affected by the acoustic noise. Previous visual VAD algorithms usually use single visual feature to detect the lip activity, such as active appearance models, optical flow or intensity variation. Based on the analysis of the weakness of each feature, we propose to combine intensity change measure and the optical flow in the mouth region, which can compensate for each other's weakness. In order to minimize the computational complexity, we develop simple measures that avoid statistical estimation or modeling. Specifically, the optical flow is the averaged motion vector of some grid regions and the intensity variation is detected by simple thresholding. To extract the mouth region, we propose a simple algorithm which first detects two eyes and uses the profile of intensity to detect the center of mouth. Experiments show that the proposed combination of two simple measures show higher detection rates for the given false positive rate than the methods that use a single feature.