The Projective Reconstruction Based on Genetic Algorithms
基于遗传算法的射影重构

In computer vision, it is called multi views 3D reconstruction for recovering both camera and object shapes from multiple images, and it is currently a topic of lively interest. A hierarchical reconstruction method had introduced in 1996. In the course of completing the hierarchical reconstruction, the projective reconstruction is the first key step, having very important effect on the precision of euclidean reconstruction. The existence methods are very efficient for simulation data, but they are not perfect for real image. Namely, they are not robust and the reliable results can only be obtained if images match accurately. In this paper, the projective reconstruction based on genetic algorithms is proposed, the projective depths are coded by using decimal system and the adaptability function is defined by a constraint of the measurement matrix rank 4. The projective depths are iteratively estimated by genetic algorithms and Singular value decomposition (SVD) so that the measurement matrix is made to be as close as possible to rank 4, and then the projective reconstruction is realized. The validity and robusticity of the proposed algorithm is confirmed by experiments.