EE Seminar: Gilboa Levy

~~Gilboa Levy, 
M.Sc. student under the supervision of Prof. Shai Avidan

Wednesday, March 11, 2015 at 15:00
Room 011, Kitot Bldg., Faculty of Engineering

Predicting Trajectories in Video Sequences by Motion Regularity

Abstract

We address the problem of object tracking in video sequences. We explore the regularity in the world to provide context for object tracking in video sequences. Many motions are repetitive and we would like to use this prior knowledge to improve our tracker. Our approach builds a database of tracklets - short tracking segments over time - and uses it to generate a prior for the tracking process. Specifically, we use the current tracklet (i.e., the track over the last few frames) as a query into a database of tracklets and use the top matches to predict where the object should move to next. This extends existing methods for adding context to tracking that only use tracklets that occur at the same time as the query tracklet. In our formulation, every tracklet that was ever recorded can be used to help predict the motion of the current object. We demonstrate the proposed method on several sequences.