• Korean Journal of Geomatics
• /
• v.2 no.1
• /
• pp.47-55
• /
• 2002

Remote sensing is the science and art of obtaining information about an object, area or phenomenon through the analysis of data acquired by a device that is not in contact with the object, area, or phenomenon under investigation./sup 1)/ EOC (Electro -Optical Camera) sensor loaded on the KOMPSAT-1 (Korea Multi- Purpose Satellite-1) performs the earth remote sensing operation. EOC can get high-resolution images of ground distance 6.6m during photographing; it is possible to get a tilt image by tilting satellite body up to 45 degrees at maximum. Accordingly, the device developed in this study enables to obtain images by photographing one pair of tilt image for the same point from two different planes. KOMPSAT-1 aims to obtain a Korean map with a scale of 1:25,000 with high resolution. The KOMPSAT-1 developed automated feature extraction system based on stereo satellite image. It overcomes the limitations of sensor and difficulties associated with preprocessing quite effectively. In case of using 6, 7 and 9 ground control points, which are evenly spread in image, with 95% of reliability for horizontal and vertical position, 3-dimensional positioning was available with accuracy of 6.0752m and 9.8274m. Therefore, less than l0m of design accuracy in KOMPSAT-1 was achieved. Also the ground position error of ortho-image, with reliability of 95%, is 17.568m. And elevation error showing 36.82m was enhanced. The reason why elevation accuracy was not good compared with the positioning accuracy used stereo image was analyzed as a problem of image matching system. Ortho-image system is advantageous if accurate altitude and production of digital elevation model are desired. The Korean map drawn on a scale of 1: 25,000 by using the new technique of KOMPSAT-1 EOC image adopted in the present study produces accurate result compared to existing mapping techniques involving high costs with less efficiency.

• Journal of Korea Multimedia Society
• /
• v.19 no.6
• /
• pp.1003-1013
• /
• 2016

This paper proposes a framework of superpixel-based vehicle detection method using plane normal vector in disparity space. We utilize two common factors for detecting vehicles: Hypothesis Generation (HG) and Hypothesis Verification (HV). At the stage of HG, we set the regions of interest (ROI) by estimating the lane, and track them to reduce computational cost of the overall processes. The image is then divided into compact superpixels, each of which is viewed as a plane composed of the normal vector in disparity space. After that, the representative normal vector is computed at a superpixel-level, which alleviates the well-known problems of conventional color-based and depth-based approaches. Based on the assumption that the central-bottom of the input image is always on the navigable region, the road and obstacle candidates are simultaneously extracted by the plane normal vectors obtained from K-means algorithm. At the stage of HV, the separated obstacle candidates are verified by employing HOG and SVM as for a feature and classifying function, respectively. To achieve this, we trained SVM classifier by HOG features of KITTI training dataset. The experimental results demonstrate that the proposed vehicle detection system outperforms the conventional HOG-based methods qualitatively and quantitatively.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2012.07a
• /
• pp.83-86
• /
• 2012

본 논문에서는 다양한 변이 추정 방식 중 영역기반(Area-based) 알고리듬과 특정기반(Feature-based) 알고리듬을 결합한 하이브리드(Hybrid) 변이추정 알고리듬을 제안한다. 제안하는 알고리듬은 Features from Accelerated Segment Test(FAST) 코너 점 추출기[2]를 이용하여 좌, 우 영상 각각의 특징 점을 추출한 후, 특징 점들의 정보를 이용한 스테레오 정함을 통해 신뢰도 높은 초기 변이지도(Disparity map)를 생생하게 된다. 그러나 생성된 초기 변이지도는 조밀하지 못하므로, 조밀한 변이 지도를 획득하기 위해 특징점이 추출된 영역에 대해서는 추정된 초기 변이 값을 이웃 픽셀과의 색 유사도를 고려하여 전파시키고 특징 점이 추출되지 않은 영역에 대해서는 이진 윈도우(Binary window)를 활용한 영역기반 변이추정 알고리듬[1]을 이용하여 변이 값을 추정한다. 이를 통해, 제안 알고리듬은 특징 기반 알고리듬에서 발생할 수 있는 보간법 문제를 해결함과 동시에 신뢰도가 높은 초기 변이지도를 사용함으로써, 영역 기반 알고리듬의 정합 오차를 줄여 신뢰도 높은 변이지도를 생생할 수 있다. 실험 결과 추정된 초기 변이지도는 ground truth와 비교 시 약 99%이상의 정확도를 보이며, 특징 점이 추출된 영역에서 기존의 영역기반 알고리듬보다 더 정확한 변이 값이 추정되었음을 확인하였다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.10
• /
• pp.4968-4986
• /
• 2017

In this paper, we primarily address the difficulty of automatic generation of a plausible depth map from a single image in an unstructured environment. The aim is to extrapolate a depth map with a more correct, rich, and distinct depth order, which is both quantitatively accurate as well as visually pleasing. Our technique, which is fundamentally based on a preexisting DepthTransfer algorithm, transfers depth information at the level of superpixels. This occurs within a framework that replaces a pixel basis with one of instance-based learning. A vital superpixels feature enhancing matching precision is posterior incorporation of predictive semantic labels into the depth extraction procedure. Finally, a modified Cross Bilateral Filter is leveraged to augment the final depth field. For training and evaluation, experiments were conducted using the Make3D Range Image Dataset and vividly demonstrate that this depth estimation method outperforms state-of-the-art methods for the correlation coefficient metric, mean log10 error and root mean squared error, and achieves comparable performance for the average relative error metric in both efficacy and computational efficiency. This approach can be utilized to automatically convert 2D images into stereo for 3D visualization, producing anaglyph images that are visually superior in realism and simultaneously more immersive.