摘要 |
A method of processing an input sequence of images to determine the positions and orientations at which the images were recorded. A subset of the input images are selected as keyframes. The error for the camera projections of images in the subset is calculated by projecting the selected features points into the images in the subset and determining the distance between positions of projected points and corresponding features in the images and changed positions for at least some of the selected points are calculated to determine the change in camera projection for the subset which reduces the calculated error. The subsets are selected sequentially from the sequence such that each image in the sequence is processed at least once. The error may be reduced using a non-linear least squares minimisation procedure. |