Robust Camera Pose Recovery Using Stochastic Geometry

The objective of three-dimensional (3-D) machine vision is to infer geometric properties (shape, dimensions) and photometric attributes (color, texture, reflectance) from a set of two-dimensional images. Such vision tasks rely on accurate camera calibration, that is, estimates of the camera's intrinsic parameters, such as focal length, principal point, and radial lens distortion, and extrinsic parameters-orientation, position, and scale relative to a fixed frame of reference. This thesis introduces methods for automatic recovery of precise extrinsic camera pose among a large set of images, assuming that accurate intrinsic parameters and rough estimates of extrinsic parameters are available. Although the motivating application is metric 3-D reconstruction of urban environments from pose-annotated hemispherical imagery, few domain-specific restrictions are required or imposed. Orientation is recovered independently of position via the detection and optimal alignment o