Three-Dimensional Reconstruction Of Geometry From Image

Three-dimensional reconstruction is a basic research question of computer graphics and computer vision. In real life,3D model reconstruction is widely used in object recognition, industrial automatic design, video games, animation, restore the original appearance of buildings and so on, while a higher reconstruction efficiency required to achieve real-time calculation. However, using software (such as 3D MAX and Maya, etc.) construct three-dimensional model by hand is a very tedious and costly work. Therefore, studying how obtain three-dimensional model directly and quickly from the real world, become a hot issue in the field.Realistic three-dimensional reconstruction is divided into active and passive methods. Active method is represented by three-dimensional scanner method. Passive method is that two-dimensional image-based three-dimensional reconstruction. Image-based 3D modeling method which has low cost, flexible and directly get the color texture, can reconstruct three-dimensional model fast and realistic.Image-based three-dimensional reconstruction is a significant work, there are automatic three-dimensional reconstruction, but reconstruction result is not satisfactory. We think that the user can interact with system, add some shape editing and increase flexibility of the system, but too many human-computer interaction will increase burden to the end user. Therefore, this work is mainly devoted to manually add the interaction of three-dimensional reconstruction of the building systems, design the corresponding algorithm to reduce end user workload.The main contributions are the following:1. Present a building reconstruction and modeling system which allows novice users to create textured polygonal models from a set of photographs of a building.2. At present, the extracted information contains too many inconsistencies, missing features, and unwanted noise to be completely reliable. Therefore, we aim at creating a semi-automatic solution which helps reduce steps in footprint drawing process and increase the accuracy of the result. 3. Exacting building footprint from point cloud. We have designed an algorithm to predict the continuation of a user-traced line segment, and provided visual cue to assist the drawing process. As the user moves the mouse, the system displays corresponding points in the image to help the user draw the first line segment.Once the first line segment is drawn, the system, given a set of adaptive parameters, uses Hough transform and least-square fitting methods to automatically deduce subsequent line segments.User can then opt to confirm or reject the deduced line segments.