http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
서재규(Jae Kyu Suhr),배광혁(Bea Kwanghyuk),정호기(Ho Gi Jung),김재희(Jaihie Kim) 대한전자공학회 2007 대한전자공학회 학술대회 Vol.2007 No.7
This paper proposes an algorithm for rejecting mismatched points (known as outliers). The proposed algorithm identifies and rejects outliers in image pairs obtained under automobile-like motions which consist of two translations and one rotation. The camera rotation is approximated to the image shift by assuming that the narrow field of lens is used. The voting method estimates the focus of expansion (FOE) while shifting one of the images. Using the properties of the FOE, the outliers are rejected while most of the inliers are retained.
컴퓨터 비전 기반 자동주차 시스템을 위한 대응점 추출 방법 비교
서재규(Jae Kyu Suhr),정호기(Ho Gi Jung),김재희(Jaihie Kim) 대한전자공학회 2007 대한전자공학회 학술대회 Vol.2007 No.11
Computer vision-based automatic parking system reconstructs 3D structure of parking space and automatically finds a free parking space. For this task, point correspondences must be found on the automobile surface. Therefore, we evaluate the methods for finding point correspondences in this paper. As a result, feature tracking method is found to be the best in terms of point correspondence number, real-time processing, and key frame selection.
Piece-wise Linear 함수 기반 스테레오 영상에서의 도로면 추정 개선 방법
서재규(Jae Kyu Suhr),정호기(Ho Gi Jung) 한국자동차공학회 2012 한국자동차공학회 학술대회 및 전시회 Vol.2012 No.11
This paper proposes a novel road surface estimation method using a piece-wise linear function and RANSAC framework. The proposed method achieves robustness against 3D points on obstacle surfaces by sampling 3D points expected to compose road surface, and it also makes the estimation procedure insensitive to stereo matching errors on textureless road regions by sequentially calculating a piece-wise linear function using a RANSAC-based robust line estimator with adaptively chosen road interval and slope angle parameters. Experimental results show that the proposed method successfully estimates road surfaces in various real world situations including complex road surface shape and severe stereo matching error.
도심지역 적용을 위한 uv-disparity 기반 장애물 검출 개선 방안
서재규(Jae Kyu Suhr),정호기(Ho Gi Jung) 한국자동차공학회 2011 한국자동차공학회 학술대회 및 전시회 Vol.2011 No.11
Obstacle detection based on v-disparity is one of the popular methods for detecting obstacles using a stereo camera. However, when applying this method to urban environments, v-disparity can severely be affected by buildings beside the roads. Therefore, this paper proposes a method to remove building surfaces from uv-disparity for enhancing the performance of vdisparity-based obstacle detection in urban environments. The proposed method first finds pixels consisting road surface and detects lanes to estimate the location of vanishing point. Then, u-disparity is divided into left and right parts according to the location of the vanishing point. Finally, building surfaces are detected and eliminated by estimating lines in each part of udisparity by using Hough transformation. Experimental results reveal that the uv-disparity with the proposed method more clearly shows obstacle regions compared to the uv-disparity without it.
서재규(Jae Kyu Suhr),정호기(Ho Gi Jung) 한국자동차공학회 2013 한국자동차공학회 부문종합 학술대회 Vol.2013 No.5
Almost all the previous automatic parking systems update relative position between ego-vehicle and target parking slot by utilizing only in-vehicle motion sensor-based odometry. However, performance of these systems can be degraded due to cumulative errors of in-vehicle motion sensor-based odometry. To overcome this drawback, this paper proposes a method that continuously tracks target parking slot position by fusing around view monitor (AVM) images and in-vehicle motion sensor-based odometry. In experiments, the proposed method successfully tracks various types of parking slot markings in spite of severe occlusions caused by ego-vehicle.
서재규(Jae Kyu Suhr),정호기(Ho Gi Jung) 한국자동차공학회 2012 한국자동차공학회 부문종합 학술대회 Vol.2012 No.5
This paper proposes a method for detecting parking slot markings in Around View Monitor (AVM) image sequences. The proposed method consists of the following four procedures: 1) Initial parking slots are detected in the current image using a hierarchical tree structure; 2) The current positions of previously detected parking slots are predicted; 3) Detected and predicted parking slots are combined; 4) Final parking slots are determined by clustering parking slots according to type and orientation. Experimental results show that this method can successfully find various parking slot marking in practical situations.
서재규(Jae Kyu Suhr),강형민(Hyung Min Kang),정호기(Ho Gi Jung) 한국자동차공학회 2012 한국자동차공학회 부문종합 학술대회 Vol.2012 No.5
This paper proposes a selection method of regions of interest (ROIs) for pedestrian detection using stereo vision. We define a ROI as a place where both driver and pedestrian are hardly see each other since this is dangerous location from a pedestrian detection viewpoint. The proposed method first finds boundary between the ground plane and obstacle based on UV-disparity, and then measures the depth change of obstacles. Finally, ROIs of pedestrian detector are selected by finding local extrema of the measured depth change. This approach can dramatically reduce the ROIs of pedestrian detection in order to decrease both false alarm and computational cost.