Abstract
This article focuses on mathematical aspects of computer vision, emphasizing the fact that many mathematically interesting facts are found through the process of designing and implementing practical engineering systems. This viewpoint is illustrated by analyzing the reliability of 3-D reconstruction from two images by stereo. First, the camera imaging is modeled as perspective projection, and the geometry of stereo is mathematically defined. It is shown that by introducing a model of image noise and applying statistical optimization, the 3-D positions of feature points are optimally reconstructed and their covariance matrices are quantitatively evaluated. Finally, related mathematically interesting topics are described from an engineering perspective.
