Dense and Sparse Optic Flows Aggregation for Accurate Motion Segmentation in Monocular Video Sequences

摘要

This paper proposes a new approach to motion segmentation in video sequences based on the aggregation of velocity fields produced by dense and sparse optic flow estimators. In the beginning, sparse optic flow information is used to identify a set of control points on moving objects. The next step relies on dense optical flow to cluster the set of control points and determine the concave hull of moving image regions. In the final step, the silhouette of these regions is extracted using active contours. The result of the proposed algorithm is a pixel-accurate motion mask that can serve as input in various scenarios ranging from surveillance systems to videoconferencing applications.