• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2012.11a
• /
• pp.11-14
• /
• 2012

본 논문에서는 키넥트 센서의 RGB영상과 깊이영상을 사용하여 얼굴을 검출하고, 검출 된 템플릿을 이용하여 얼굴을 추적하는 방법을 제안한다. 얼굴검출은 기본적으로 기존의 Adaboost 방법을 사용하나, 깊이정보와 피부색을 사용하여 탐색영역을 최대한 축소하여 수행시간 및 오검출율을 줄였다. 그리고 얼굴추적은 깊이정보를 이용하여 템플릿의 크기, 탐색영역을 조정하였다. 또한, RGB영상보다 조명변화에 강한 깊이영상을 이용하여 효율적인 템플릿 매칭을 하였다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.17 no.5
• /
• pp.1119-1128
• /
• 2013

In this paper, we propose an algorithm to detect and track the human face with a GPU-based high speed. Basically the detection algorithm uses the existing Adaboost algorithm but the search area is dramatically reduced by detecting movement and skin color region. Differently from detection process, tracking algorithm uses only depth information. Basically it uses a template matching method such that it searches a matched block to the template. Also, In order to fast track the face, it was computed in parallel using GPU about the template matching. Experimental results show that the GPU speed when compared with the CPU has been increased to up to 49 times.

• Proceedings of the Korean Information Science Society Conference
• /
• 2005.11b
• /
• pp.844-846
• /
• 2005

본 논문은 시점에 독립적인 제스처 인식을 위하여 볼륨 모션 템플릿을 제안한다. 기존 제스처 연구에서 시점 문제와 행동 속도의 편차는 중요하면서도 어려운 문제이다. 첫째, 시점 문제는 하나의 단안 카메라나 스테레오 카메라를 이용하는 단방향 카메라 환경에서 발생하며 해결하기 어려운 문제이다. 모든 시점에서 학습시켜야 하는 기존 연구의 단점을 해결하기 위해, 다양한 시점입력에 독립적으로 인식을 할 수 있는 볼륨 모션 템플릿을 제안한다. 볼륨 모션 템플릿은 깊이 정보와 모션의 방향성 통해 최적의 가상 시점을 제공한다. 또한 볼륨 모션 템플릿을 이용하여 시스템의 신뢰성과 확장성 또한 개선하였다. 두 번째, 제스처가 발생 시마다 생기는 속도의 편차 문제이다. 입력 제스처의 시간-정규화를 통해 해결할 수 있는데, 시간 정보 대신 모션 량을 사용하여 이를 해결하였다. 볼륨 모션 템플릿을 이용하여 다양한 시점 입력에 대해 실험하였고, 기존 모션 히스토리 이미지와 비교하여 시점에 독립적인 결과를 얻었다.

• Journal of Broadcast Engineering
• /
• v.18 no.3
• /
• pp.349-361
• /
• 2013

This paper proposes a fast face tracking method with only depth information. It is basically a template matching method, but it uses a early termination scheme and a sparse search scheme to reduce the execution time to solve the problem of a template matching method, large execution time. Also a refinement process with the neighboring pixels is incorporated to alleviate the tracking error. The depth change of the face being tracked is compensated by predicting the depth of the face and resizing the template. Also the search area is adjusted on the basis of the resized template. With home-made test sequences, the parameters to be used in face tracking are determined empirically. Then the proposed algorithm and the extracted parameters are applied to the other home-made test sequences and a MPEG multi-view test sequence. The experimental results showed that the average tracking error and the execution time for the home-made sequences by Kinect ($640{\times}480$) were about 3% and 2.45ms, while the MPEG test sequence ($1024{\times}768$) showed about 1% of tracking error and 7.46ms of execution time.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2013.11a
• /
• pp.120-123
• /
• 2013

본 논문에서는 키넥트 센서의 RGB영상을 이용하여 얼굴을 검출하고 검출된 영역의 깊이정보를 템플릿으로 사용하여 다수개의 얼굴을 추적하는 방법을 제안한다. 이 논문은 [1]의 단일 얼굴 추적방법을 다수의 얼굴을 추적하도록 확장한 것이다. 다수의 얼굴추적을 실시간으로 처리하기 위하여 영상을 down sampling 하여 사용한다. 얼굴 검출은 기본적으로 기존의 Adaboost 방법을 사용하나, 피부색만을 이용, 탐색영역을 최대한 축소하여 수행 시간 및 오검출율을 줄인다. 얼굴추적은 깊이정보를 템플릿으로 하며, 깊이값에 따라 크기, 탐색영역을 조정하고, 또한 일정 프레임마다 얼굴을 검출하며 겹침, 새로 나타남, 영상 밖으로 사라짐 등의 얼굴추적 시 발생하는 문제를 해결한다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2012.07a
• /
• pp.68-70
• /
• 2012

본 논문에서는 컬러영상과 깊이영상을 이용하여 여러 명의 얼굴을 검출하고 추적하는 알고리즘을 제안한다. 제안하는 알고리즘은 얼굴 검출부와 추적부로 나뉘어져 있으며, 얼굴 검출 방법은 기존의 Adaboost를 이용하지만, 속도 개선을 위해 깊이정보와 컬러정보를 이용하여 탐색영역을 얼굴이 존재하는 영역으로 제한하여 얼굴은 검출한다. 얼굴 추적 방법은 템플릿 매칭 방법과 나선형 탐색방법을 사용하며, 그리고 조기 종료 기법을 사용하여 수행시간을 줄였다.

• Journal of Broadcast Engineering
• /
• v.17 no.3
• /
• pp.437-446
• /
• 2012

A face detection algorithms using two-dimensional (2-D) intensity or color images have been studied for decades. Recently, with the development of low-cost range sensor, three-dimensional (3-D) information (i.e., depth image that represents the distance between a camera and objects) can be easily used to reliably extract facial features. Most people have a similar pattern of 3-D facial structure. This paper proposes a face detection method using intensity and depth images. At first, adaboost algorithm using intensity image classifies face and nonface candidate regions. Each candidate region is divided into $5{\times}5$ blocks and depth values are averaged in each block. Then, $5{\times}5$ block rank pattern is constructed by sorting block averages of depth values. Finally, candidate regions are classified as face and nonface regions by matching the constructed depth map based block rank patterns and a template pattern that is generated from training data set. For template matching, the $5{\times}5$ template block rank pattern is prior constructed by averaging block ranks using training data set. The proposed algorithm is tested on real images obtained by Kinect range sensor. Experimental results show that the proposed algorithm effectively eliminates most false positives with true positives well preserved.

• Journal of the Korea Computer Graphics Society
• /
• v.20 no.2
• /
• pp.1-11
• /
• 2014

The quality of 2D-to-3D conversion depends on the accuracy of the assigned depth to scene objects. Manual depth painting for given objects is labor intensive as each frame is painted. Specifically, a human is one of the most challenging objects for a high-quality conversion, as a human body is an articulated figure and has many degrees of freedom (DOF). In addition, various styles of clothes, accessories, and hair create a very complex silhouette around the 2D human object. We propose an efficient method to estimate visually pleasing depths of a human at every frame in a monocular video. First, a 3D template model is matched to a person in a monocular video with a small number of specified user correspondences. Our pose estimation with sequential joint angular constraints reproduces a various range of human motions (i.e., spine bending) by allowing the utilization of a fully skinned 3D model with a large number of joints and DOFs. The initial depth of the 2D object in the video is assigned from the matched results, and then propagated toward areas where the depth is missing to produce a complete depth map. For the effective handling of the complex silhouettes and appearances, we introduce a partial depth propagation method based on color segmentation to ensure the detail of the results. We compared the result and depth maps painted by experienced artists. The comparison shows that our method produces viable depth maps of humans in monocular videos efficiently.

• The Journal of Korean Academy of Prosthodontics
• /
• v.52 no.4
• /
• pp.366-375
• /
• 2014

Surgical guide not only provide diagnosis and treatment plan, but even location and direction of implantation. Surgical guide could be divided into non-limited design, partially limited design, and completely limited design. Partially limited design is easily manufactured and inexpensive but less accuracy, compared to completely limited design. From this approach, partially limited design may be particularly effective in patients who present with a single missing tooth or partially edentulous teeth. Completely limited design is anatomically accuracy, esthetical and functional, optimized treatment for prosthetic and biomechanical perspective, and also minimizes discomfort for post-treatment. The purpose of this study is to review previous studies of various surgical guides and applying in clinic.

• The Journal of the Korea Contents Association
• /
• v.14 no.12
• /
• pp.545-554
• /
• 2014

In this paper, we propose online Chinese character learning machine with a depth camera, where a system presents a Chinese character on a screen and a user is supposed to draw the presented Chinese character by his or her hand gesture. We develop the hand tracking method and suggest the adjusted chain code to represent constituent strokes of a Chinese character. For hand tracking, a fingertip is detected and verified. The adjusted chain code is designed to contain the information on order and relative length of each constituent stroke as well as the information on the directional variation of sample points. Such information is very efficient for a real-time match process and checking incorrectly drawn parts of a stroke.