Long-range video motion estimation using point trajectories

Abstract

This thesis describes a new approach to video motion estimation, in which motion is represented using a set of particles. Each particle is an image point sample with a long-duration trajectory and other properties. To optimize these particles, we measure point-based matching along the particle trajectories and distortion between the particles. The resulting motion representation is useful for a variety of applications and differs from optical flow, feature tracking, and parametric or layer-based models. We demonstrate the algorithm on challenging real-world videos that include complex scene geometry, multiple types of occlusion, regions with low texture, and non-rigid deformation.