Ordering random object poses

Complete or partial three-dimensional reconstruction of objects from multiple angle-views, or poses, is important in several applications such as photogrammetry, machine vision, and computer-aided design. Knowledge of the pose angles and their proper ordering are required for accurate reconstruction. When these multiple angle images are acquired in random order and the angle of view information is not available the poses have to be put into proper order. This work presents an approach based on principal component analysis (PCA) for automatic ordering of random object poses. A measure based on local curvature and correlation of the estimated pose trajectory in a multidimensional manifold is also developed to assess confidence in the ordering. In addition to providing a degree of confidence for pose ordering with single cameras, this measure enhances the pose estimation accuracy in double and multiple camera systems by providing a basis for camera selection for different poses. The paper presents theoretical development and experimental results.