부산대학교 도서관

We present an algorithm that efficiently tracks a predefined set of landmark points in a time sequence of images. The algorithm iteratively optimizes the correspondences between the point measurements in the images, while allowing for spurious and missing point measurements. This trajectory based approach hypothesizes missing points by interpolation. Spurious measurements are either left our because they do not form the optimal correspondences or are removed afterwards if they have the smoothness or other constraint exceed its predetermined maximum.