Towards Longer Long-Range Motion Trajectories

Although dense, long-rage, motion trajectories are a prominent representation of motion in videos, there is still no good solution for constructing dense motion tracks in a truly long-rage fashion. Ideally, we would want every scene feature that appears in multiple, not necessarily contiguous, parts of the sequence to be associated with the same motion track. Despite this reasonable and clearly stated objective, there has been surprisingly little work on general-purpose algorithms that can accomplish that task. State-of-the-art dense motion trackers process the sequence incrementally in a frame-by-frame manner, and associate, by design, features that disappear and reappear in the video, with different tracks, thereby losing important information of the long-term motion signal. In this paper, we propose a novel divide and conquer approach to long-range motion estimation. Given a long video or image sequence, we first produce high-accuracy local track estimates, or tracklets, and later propagate them into a global solution, while incorporating information from throughout the video. Tracklets are computed using state-of-the-art motion trackers [2, 3] that have become quite accurate for short sequences as demonstrated by standard evaluations. Our algorithm then constructs the long-range tracks by linking the short tracks in an optimal manner. This induces a combinatorial matching problem that we solve simultaneously for all tracklets in the sequence. The main contributions of this paper are: (a) a novel divide-andconquer style algorithm for constructing dense, long-rage motion tracks from a single monocular video, and (b) Novel criteria for evaluating longrange tracking results with and without ground-truth motion trajectory data. We evaluate our approach on a set of synthetic and natural videos, and explore the utilization of long-range tracks for action recognition.