February 2005
·
10 Reads
·
3 Citations
A novel method for the three dimensional (3D) reconstruction from two uncalibrated images was described. The camera intrinsic parameters can be estimated linearly by using three pairs of corresponding vanishing points from the three mutual orthogonal space directions, then the camera motion parameters between two views can be estimated from three groups of lines that are mutually orthogonal in 3D space. After calculating the camera projection matrices, the coordinates in 3D space can be calculated using triangular measurement. Comparing with the active camera self-calibration method using three mutual orthogonal translations, the camera orthogonal motion constraints are transformed into the orthogonal constraints of the spatial structure in our approach, so that it is effective to be realized with higher adaptability and the 3D models of the scene can be recovered from 2D images taken by an uncalibrated handhold digital camera. This approach has been applied to the real images of architectural scenes, and a 3D model of the building was reconstructed with better performance. The new images generated from the reconstructed 3D model for the new viewpoints are consistent with the perception of the real scene, and the measure error of the plane angle between the reconstructed 3D model and the real scene is within 1.5%-2.6%.