Motion segmentation by model-based clustering of incomplete trajectories

摘要: In this paper, we present a framework for visual object tracking based on clustering trajectories of image key points extracted from video. The main contribution our method is that the are automatically video sequence and they provided directly to model-based approach. most other methodologies, latter constitutes difficult part since resulting feature have short duration, as disappear reappear due occlusion, illumination, viewpoint changes noise. We here sparse, translation invariant regression mixture model variable length. overall scheme converted into Maximum A Posteriori approach, where Expectation-Maximization (EM) algorithm used estimating parameters. proposed detects different objects in input by assigning each trajectory cluster, simultaneously provides motion all objects. Numerical results demonstrate ability offer more accurate robust solution comparison with mean shift tracker, especially cases occlusions.