• Journal of the Institute of Convergence Signal Processing
• /
• v.14 no.1
• /
• pp.33-38
• /
• 2013

In this paper, we propose an algorithm to track multi-small targets in infrared image sequences in case of dissipation or creation of targets by using the background estimation filter, Kahnan filter and mean shift algorithm. We detect target candidates in a still image by subtracting an original image from an background estimation image, and we track multi-targets by using Kahnan filter and target selection. At last, we adjust specific position of targets by using mean shift algorithm In the experiments, we compare the performance of each background estimation filters, and verified that proposed algorithm exhibits better performance compared to classic methods.

• Journal of KIISE:Computer Systems and Theory
• /
• v.33 no.6
• /
• pp.354-362
• /
• 2006

This paper suggests a method for generating high qualify stereo video avatar to support visual communication in a CAVE$^{TM}$-like system. In such a system because of frequent change of light projected onto screens around user, it is not easy to extract user silhouette robustly, which is an essential step to generate a video avatar. In this study, we use an infrared reflective image acquired by a grayscale camera with a longpass filter so that the change of visible light on a screen is blocked to extract robust user silhouette. In addition, using two color cameras positioned at a distance of a binocular disparity of human eyes, we acquire two stereo images of the user for fast generation and stereoscopic display of a high quality video avatar without 3D reconstruction. We also suggest a fitting algorithm of a silhouette mask on an infrared reflective image into an acquired color image to remove background. Generated stereo images of a video avatar are texture mapped into a plane in virtual world and can be displayed in stereoscopic using frame sequential stereo method. Suggested method have advantages that it generates high quality video avatar taster than 3D approach and it gives stereoscopic feeling to a user 2D based approach can not provide.

• 한국HCI학회:학술대회논문집
• /
• 2006.02a
• /
• pp.1188-1193
• /
• 2006

본 논문에서는 동적 객체의 3 차원 정보를 표현하는 깊이 영상의 노이즈 필터링 방법을 제안한다. 실제 객체의 동적인 3 차원 정보는 적외선 깊이 센서가 장착된 깊이 비디오 카메라를 이용하여 실시간으로 획득되며, 일련의 깊이 영상, 즉 깊이 비디오(depth video)로 표현될 수 있다. 하지만 측정환경의 조명조건, 객체의 반사속성, 카메라의 시스템 오차 등으로 인해 깊이 영상에는 고주파 성분의 노이즈가 발생하게 된다. 이를 효과적으로 제거하기 위해 깊이 영상기반의 모델링 기법(depth image-based modeling)을 이용한 3 차원 메쉬 모델링을 수행한다. 생성된 3 차원 메쉬 모델은 깊이 영상의 노이즈로 인해 경계 영역과 형상 내부 영역에 심각한 형상 오차를 가진다. 경계 영역의 오차를 제거하기 위해 깊이 영상으로부터 경계 영역을 추출하고, 가까운 순서로 정렬한 후 angular deviation 을 이용하여 불필요하게 중복된 점들을 제거한다. 그리고 나서 2 차원 가우시안 스무딩 기법을 적용하여 부드러운 경계영역을 생성한다. 형상 내부에 대해서는 경계영역에 제약조건을 주고 3 차원 가우시안 스무딩 기법을 적용하여 전체적으로 부드러운 형상을 생성한다. 최종적으로 스무딩된 3 차원 메쉬모델을 렌더링할 때, 깊이 버퍼에 있는 정규화된 깊이 값들을 추출하여 원래 깊이 영상과 동일한 깊이 영역을 가지도록 저장함으로서 전역적으로 연속적이면서 부드러운 깊이 영상을 생성할 수 있다. 제안된 방법에 의해 노이즈가 제거된 깊이 영상을 이용하여 고품질의 영상기반 렌더링이나 깊이 비디오 기반의 햅틱 렌더링에 적용할 수 있다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.37 no.4A
• /
• pp.239-249
• /
• 2012

Recently, the virtual view generation method using depth data is employed to support the advanced stereoscopic and auto-stereoscopic displays. Although depth data is invisible to user at 3D video rendering, its accuracy is very important since it determines the quality of generated virtual view image. Many works are related to such depth enhancement exploiting a time-of-flight (TOF) camera. In this paper, we propose a fast 3D scene capturing system using one TOF camera at center and two high-resolution cameras at both sides. Since we need two depth data for both color cameras, we obtain two views' depth data from the center using the 3D warping technique. Holes in warped depth maps are filled by referring to the surrounded background depth values. In order to reduce mismatches of object boundaries between the depth and color images, we used the joint bilateral filter on the warped depth data. Finally, using two color images and depth maps, we generated 10 additional intermediate images. To realize fast capturing system, we implemented the proposed system using multi-threading technique. Experimental results show that the proposed capturing system captured two viewpoints' color and depth videos in real-time and generated 10 additional views at 7 fps.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.22 no.4
• /
• pp.494-499
• /
• 2012

In this paper, a new method for detecting various objects that can be risks to safety navigation in sea environment is proposed. By analysing Infrared(IR) images obtained from various sea environments, we could find out that object regions include both horizontal and vertical direction edges while background regions of sea surface mainly include vertical direction edges. Therefore, we present an approach to detecting object regions considering horizontal and vertical edges. To this end, in the first step, image enhancement is performed by suppressing noises such as sea glint and complex clutters using a statistical filter. In the second step, a horizontal edge map and a vertical edge map are generated by 1-D Discrete Cosine Transform technique. Then, a combined map integrating the horizontal and the vertical edge maps is generated. In the third step, candidate object regions are detected by a adaptive thresholding method. Finally, exact object regions are extracted by eliminating background and clutter regions based on morphological operation.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.6
• /
• pp.1-8
• /
• 2021

In this paper, we propose reinforced VGG style network structure for low performance embedded system to classify low resolution infrared image. The combination of reinforced VGG style network structure and global average pooling makes lower computational complexity and higher accuracy. The proposed method classify the synthesize image which have 9 class 3,723,328ea images made from OKTAL-SE tool. The reinforced VGG style network structure composed of 4 filters on input and 16 filters on output from max pooling layer shows about 34% lower computational complexity and about 2.4% higher accuracy then the first parameter minimized network structure made for embedded system composed of 8 filters on input and 8 filters on output from max pooling layer. Finally we get 96.1% accuracy model. Additionally we confirmed the about 31% lower inference lead time in ported C code.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38C no.5
• /
• pp.409-419
• /
• 2013

Kinect released by Microsoft in November 2010 is a motion sensing camera in xbox360 and gives depth and color images. However, Kinect camera also generates holes and noise around object boundaries in the obtained images because it uses infrared pattern. Also, boundary flickering phenomenon occurs. Therefore, we propose a real-time virtual-view video synthesis algorithm which results in a high-quality virtual view by solving these problems. In the proposed algorithm, holes around the boundary are filled by using the joint bilateral filter. Color image is converted into intensity image and then flickering pixels are searched by analyzing the variation of intensity and depth images. Finally, boundary flickering phenomenon can be reduced by converting values of flickering pixels into the maximum pixel value of a previous depth image and virtual views are generated by applying 3D warping technique. Holes existing on regions that are not part of occlusion region are also filled with a center pixel value of the highest reliability block after the final block reliability is calculated by using a block based gradient searching algorithm with block reliability. The experimental results show that the proposed algorithm generated the virtual view image in real-time.

• Proceedings of the Korean Information Science Society Conference
• /
• 2005.07b
• /
• pp.814-816
• /
• 2005

본 연구에서는 영상획득 조건을 충분히 만족시키지 못하는 몰입형 디스플레이 환경에서 다수의 사용자 머리위치를 탐지하는 방법을 제안한다. 본 기법은 몰입형 가상환경에서 적외선 반사영상을 획득하고 그로부터 배경을 제거함으로써 얻어진 전경(foreground) 영역으로부터 프로젝션 히스토그램을 생성하고 사용자 실루엣을 추출하게 된다. 모든 사용자의 머리는 다각형 근사된 실루엣과 프로젝션 히스토그램에 기반하여 탐지된다. 또한 향후 몰입형 가상환경에서의 다중 사용자 트래킹을 지원하기 위해 스테레오 영상에서 탐지된 머리를 기준으로 탐색영역을 정의, 대응점을 결정하고 그에 기반하여 각 사용자 머리의 3차원 위치를 계산하였다.

• Journal of Korea Multimedia Society
• /
• v.16 no.6
• /
• pp.746-755
• /
• 2013

This paper presents a three dimensional bilateral filter detecting target trajectory, extracting spatial target information using two dimensional bilateral filter and temporal target information using one dimensional bilateral filter. In order to discriminate edge pixel with flat background and target region spatially and temporally, spatial and temporal variance are used for an image and temporal profile. With this procedure, background and background profile are predicted without original target through two dimensional and one dimensional bilateral filter. Finally, using spatially predicted background and temporally predicted background profile, small target can be detected. For comparison of existing target detection methods and the proposed method, the receiver operating characteristics (ROC) is used in experimental results. Experimental results show that the proposed method has superior target detection rate and lower false alarm rate.

• Journal of Digital Contents Society
• /
• v.19 no.6
• /
• pp.1213-1218
• /
• 2018

Recently introduced Mask R - CNN presents a conceptually simple, flexible, general framework for instance segmentation of objects. In this paper, we propose an algorithm for efficiently searching objects of images, while creating a segmentation mask of heat generation part for an instance which is a heating element in a heat sensed image acquired from a thermal infrared camera. This method called a mask R - CNN is an algorithm that extends Faster R - CNN by adding a branch for predicting an object mask in parallel with an existing branch for recognition of a bounding box. The mask R - CNN is added to the high - speed R - CNN which training is easy and fast to execute. Also, it is easy to generalize the mask R - CNN to other tasks. In this research, we propose an infrared image detection algorithm based on R - CNN and detect heating elements which can not be distinguished by RGB images. As a result of the experiment, a heat-generating object which can not be discriminated from Mask R-CNN was detected normally.