Pose and position estimations of regular shapes in monocular images

• Published in: 2009 International Conference on Machine Learning and Cybernetics Vol. 4; pp. 2463 - 2469
• Main Authors: Yuan Ren, Hui Wei
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2009
• Subjects: Computer science; Computer vision; Cybernetics; Digital cameras; Digital images; Effective focal length; Electronic mail; Machine learning; Machine learning algorithms; Perspective rectangle; Pose; Position; Prior knowledge; Robot vision systems; Shape
• ISBN: 9781424437023; 1424437024
• ISSN: 2160-133X
• DOI: 10.1109/ICMLC.2009.5212237

This paper presents approaches for pose and position estimations of two-dimensional regular shapes including rectangles and circles as well as particular kinds of triangles in monocular images. Part of this problem has been settled with parameters of digital cameras calibrated, which is not infeasible for all encountered images. Take the formulation of an image as perspective projection and the effective focal length of the cameras behind can be uniquely determined by only one general perspective rectangle in the image. Then with no additional condition required, the poses of those shapes concerned here with respect to the camera coordinate system can be calculated. Further, their positions can also be estimated with reasonable prior knowledge about lengths or areas. Experiment shows results desirably according with both the real fact and people's perceptions. Compared with other methods, ours are not just less complex, numerically solvable and intuitionistic but robust to errors in coordinates and changes made to images.