作者机构：
[Li, Chang; Liu, Pengcheng] Huazhong Normal Univ, Coll Urban & Environm Sci, Wuhan 430079, Peoples R China.;[Yang, Ling] Henan Univ, China Australia Cooperat Res Ctr Geoinformat Anal, Coll Environ & Planning, Kaifeng 475004, Peoples R China.;[Hu, Min] Peoples Liberat Army Commun Command Acad, Wuhan 430010, Peoples R China.;[Hu, Min] China Univ Geosci, Sch Econom & Management, Wuhan 430074, Peoples R China.
通讯机构：
[Li, Chang] H;Huazhong Normal Univ, Coll Urban & Environm Sci, Wuhan 430079, Peoples R China.
会议名称：
International Conference on Advanced Measurement and Test (AMT 2010)
会议时间：
MAY 15-16, 2010
会议地点：
Sanya, PEOPLES R CHINA
会议主办单位：
[Li, Chang;Liu, Pengcheng] Huazhong Normal Univ, Coll Urban & Environm Sci, Wuhan 430079, Peoples R China.^[Yang, Ling] Henan Univ, China Australia Cooperat Res Ctr Geoinformat Anal, Coll Environ & Planning, Kaifeng 475004, Peoples R China.^[Hu, Min] Peoples Liberat Army Commun Command Acad, Wuhan 430010, Peoples R China.^[Hu, Min] China Univ Geosci, Sch Econom & Management, Wuhan 430074, Peoples R China.
会议论文集名称：
Key Engineering Materials
关键词：
vanishing point;line clustering;RANSAC;condition adjustment with parameters;error ellipse
摘要：
In close-range digital photogrammetry and computer vision, a major challenge is the automation of 3D reconstruction from 2D-images. And single image calibration is a fundamental task in these areas for research. It is known that camera parameters can be recovered by the vanishing points of three orthogonal directions. However, three reliable and well-distributed vanishing points are not always available. Therefore, how to estimate the error of two vanishing points is very significant for us to analyze the precision of camera calibration. New methods for vanishing point detection and error estimation are presented, which can be illustrated as follows. Firstly, the line clustering, which parallel to object lines and correspond to the vanishing points, are detected based on RANSAC (Random Sample Consensus). Secondly, "condition adjustment with parameters" is utilized to estimate a nonlinear error equation. Thirdly, the error of vanishing point is expressed by error ellipse that is derived by co-factor matrix according to adjustment principle. Finally, experimental results of vanishing points coordinates and their errors are shown and analyzed.
作者机构：
[Li, Chang; Liu, Pengcheng] Huazhong Normal Univ, Coll Urban & Environm Sci, Wuhan 430079, Peoples R China.;[Hu, Min] Peoples Liberat Army Commun Command Acad, Wuhan 430010, Peoples R China.;[Hu, Min] China Univ Geosci, Sch Econom & Management, Wuhan 430074, Peoples R China.
通讯机构：
[Li, Chang] H;Huazhong Normal Univ, Coll Urban & Environm Sci, Wuhan 430079, Peoples R China.
会议名称：
International Conference on Advanced Measurement and Test (AMT 2010)
会议时间：
MAY 15-16, 2010
会议地点：
Sanya, PEOPLES R CHINA
会议主办单位：
[Li, Chang;Liu, Pengcheng] Huazhong Normal Univ, Coll Urban & Environm Sci, Wuhan 430079, Peoples R China.^[Hu, Min] Peoples Liberat Army Commun Command Acad, Wuhan 430010, Peoples R China.^[Hu, Min] China Univ Geosci, Sch Econom & Management, Wuhan 430074, Peoples R China.
会议论文集名称：
Key Engineering Materials
关键词：
line extraction;Wallis filtering;LoG;Canny;Hough transform;perceptual organization;LSTM;aerial image
摘要：
Line and plane feature can provide more information than point feature, thus 3D reconstruction based on high-level features, such as line and plane, is an important development trend in Digital Photogrammetry and Computer Vision. Several methods for extracting straight line are researched, and the main procedures can be introduced as follows. Firstly, image is preprocessed by Wallis filtering that is used to enhance the image contrast and reduce the noise, so it is easy to extract more lines. Secondly, Laplacian of Gaussian operator (LoG) and Canny operator algorithms are compared to locate the edge by detecting discontinuity variation in image. Thirdly, Hough transform or perceptual organization based on hypothesis testing are compared and tested for combining and fitting fractured short line segments into a whole line. Lastly, the least square template matching algorithm (LSTM) is done to get higher precise (sub-pixel) located lines. In the experiment, different algorithms for straight line extraction of aerial images are realized and compared, in order to faster achieve richer and higher accurate straight line information, which can pave the way of image understanding and image matching.
