Interactive player tracking for videos in American football Public Deposited

http://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/ww72bg034

Descriptions

Attribute NameValues
Creator
Abstract or Summary
  • This thesis presents an interactive software tool for tracking a moving object in a video. In particular, we focus on the problem of tracking a player in American football videos. Object tracking is one of the fundamental problems in computer vision. It is one of the most important components in numerous applications of computer vision. For our domain of American football videos, there are no existing trackers which are sufficiently robust to accurately track a player, due to many challenges (e.g. occlusion, camera motion). Our interactive software tool for tracking is aimed at improving the performance of a tracker through human corrections of tracking errors. The literature presents a number of similar interactive tools for tracking. However they are not suitable for our focus domain, because they lack in one aspect or another. Some are not real time, some require sanitized settings, some make restrictive assumptions, while some cannot handle long videos. We use the state-of-the-art method called Online Multiple Instance Learning (OMIL) tracker as our base tracker. While tracking a player, our interactive tool allows the user to tell the tracker the exact location and size of the player. Given this user feedback, the system updates the player's trajectory and places a bounding box surrounding the player in each frame. To ensure smoothness in trajectory, we use the spline interpolation to interpolate the player's trajectory between the user-specified position and the last position estimated as correct. In addition, we estimate the player's acceleration and use it for placing the bounding box along the interpolated trajectory. All this is done at nearly real time. Our large scale experiments on 355 American football videos demonstrate the effectiveness of our tracking tool. We annotated 355 videos with ground truth trajectories for a set of players of interest. We categorized tracking tasks according to player types so as to provide the user with an estimate of the number of interactions required for getting the expected degree of accuracy. A comparison of tracking results with and without our interactive tool demonstrates the increase in accuracy from 62% to 88% on average in the former case on 355 videos. We also demonstrate that the tool is not tracker dependent by testing it with another tracker called Particle Filtering.
Resource Type
Date Available
Date Copyright
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Academic Affiliation
Non-Academic Affiliation
Keyword
Subject
Rights Statement
Peer Reviewed
Language
Replaces
Additional Information
  • description.provenance : Approved for entry into archive by Julie Kurtz(julie.kurtz@oregonstate.edu) on 2014-06-07T21:44:34Z (GMT) No. of bitstreams: 1 BawaskarAmitP2014.pdf: 1152598 bytes, checksum: 4ecd5d03abc4ca253f1b00f268280e0b (MD5)
  • description.provenance : Submitted by Amit Bawaskar (bawaskaa@onid.orst.edu) on 2014-06-03T01:35:22Z No. of bitstreams: 1 BawaskarAmitP2014.pdf: 1152598 bytes, checksum: 4ecd5d03abc4ca253f1b00f268280e0b (MD5)
  • description.provenance : Made available in DSpace on 2014-06-17T21:08:23Z (GMT). No. of bitstreams: 1 BawaskarAmitP2014.pdf: 1152598 bytes, checksum: 4ecd5d03abc4ca253f1b00f268280e0b (MD5) Previous issue date: 2014-05-30
  • description.provenance : Approved for entry into archive by Laura Wilson(laura.wilson@oregonstate.edu) on 2014-06-17T21:08:23Z (GMT) No. of bitstreams: 1 BawaskarAmitP2014.pdf: 1152598 bytes, checksum: 4ecd5d03abc4ca253f1b00f268280e0b (MD5)

Relationships

Parents:

This work has no parents.

Last modified

Downloadable Content

Download PDF

Items