Research On Camera Self-calibration And Modeling Based On Geometric Features

3D reconstruction becomes the hot and difficult research direction in Computer Vision, Artificial Intelligence, Virtual Reality and other visualization fields. It is also one of the major challenges in basic and application research. With the rapid development in 3D reconstruction, Image-Based Modeling as a novel modeling method is coming out. Image-Based Modeling which combines many results in Computer Graphics, Computer Vision, Photogrammetry, Projective geometry has drawn increasing attentions in recent decades. The biggest merit of it is that posture and texture information can be directly extracted from images to reconstruct geometric models with photo-reality simply and rapidly. The research of this dissertation belongs to the field of Image-Based Modeling, and aims at applying camera self-calibration and object reconstruction based on multi-images to IBM. The main contents are as follows:1) In using vanishing point for calibration aspect, firstly, the dissertation shows constraint relationships between the pixel coordinates of vanishing points and camera matrix. Secondly, it constrains camera matrix which can be enhanced according to geometric restriction of vanishing points. Finally, camera matrix which lacks a Metric or Euclid transform can acquired and then decompose this matrix to determine intrinsic and extrinsic parameters.2) In using parallelogram for calibration aspect, every point's relative depth rate is calculated in terms of the parallelogram geometry relationship. Meanwhile, every point's analytical expression in camera coordinates can be acquired. Based on the analytical expression the focal length and other camera parameters can be drawn in terms of series functions.3) In object 3D reconstruction aspect, this dissertation shows a brand new method which reconstruct models from multi-images. At first, it use the properties of vanishing points for calculating position of camera center in world coordinates. In addition, model posture and position can be acquired by using geometry restriction of the camera. At last, the texture from images is extracted and rectified.