3D Information Acquisition from Rectangles and Circles in Single Images

This paper presents a monocular approach of partially recovering three-dimensional information from rectangles and circles with proper prior knowledge. Taking image-formation process as projective transform, we first work out the focal length of an image with a common perspective rectangle and further obtain normal vectors of the supporting planes of rectangles and circles and relative depths of each point on the planes with respect to camera coordinate system. The absolute depths and other information about lengths and distances are also acquirable with reasonable estimation or measurement of side lengths and radius. The experimental results suggest good feasibility and acceptable precision of our approach.