Full Record

New Search | Similar Records

Author
Title Offensive Direction Inference in Real-World Football Video
URL
Publication Date
Date Available
Date Accessioned
Degree MS
Discipline/Department Computer Science
Degree Level masters
University/Publisher Oregon State University
Abstract Automatic analysis of American football videos can help teams develop strategies and extract patterns with less human effort. In this work, we focus on the problem of automatically determining which team is on offense/defense, which is an important subproblem for higher-level analysis. While seemingly mundane, this problem is quite challenging when the source of football video is relatively unconstrained, which is the situation we face. Our football videos are collected from a web-service used by more than 13, 000 high school, college, and professional football teams. The videos display huge variation in camera viewpoint, lighting conditions, football field properties, camera work, among many other factors. These factors taken together make standard off-the-shelf computer vision algorithms ineffective. The main contribution of this thesis is to design and evaluate two novel approaches for the offense/defense classification problem from raw video, which make minimal assumptions about the video properties. Our empirical evaluation on approximately 1200 videos of football plays from 10 diverse games validate the effectiveness of the approaches and highlight their differences.
Subjects/Keywords American football video; Computer vision
Contributors Fern, Alan (advisor); Bailey, Mike (committee member)
Language en
Rights Attribution-NonCommercial 3.0 United States
http://creativecommons.org/licenses/by-nc/3.0/us/ [Always confirm rights and permissions with the source record.]
Country of Publication us
Record ID handle:1957/56218
Repository oregonstate
Date Indexed 2017-03-17
Grantor Oregon State University
Issued Date 2015-06-10 00:00:00
Note [] Graduation date: 2015; [peerreview] no;

Sample Search Hits | Sample Images | Cited Works

…tivity recognition applications, such as security surveillance, patient monitoring and human-computer interactions. However, the activity recognition in sports domain is still under-served. American football is the most popular sport in the United States…

…The game planning of American football teams needs a lot of annotation and analysis of videos of their own and opponent games. However, the existing services for football game planning still involve large amount of manual management, annotation, and…

…Background and Challenges American football video analysis is conducted around the concept of football plays. A play is a video clip lasting approximately 10 to 30 seconds. One football game usually contains a sequence of approximately 150 plays. One video…

…x5B;3] or random forest [30] can be trained using the computed representations of the labeled training set. 2.2 American Football Video Analysis Sports video analysis is one of these applications of activity recognition that focuses on…

…sports domain. In our work, we mainly focus on the video analysis of one specific sport­ 6 American football. Most of the work on American football analysis focuses on either classifying different types of plays or recognizing football formations. In…

…x5B;3], a framework for automatic recognition of offensive team formations in American football plays is proposed. As one stage of this framework, a method based on the spatial distribution of players is used to infer the offensive direction(…

…see more details in Section 3.3). Siddiquie et al. [33] come up with a learning based method under discriminative feature selection framework for recognizing plays in American football games. In their feature selection framework, the…

…the players’ trajectories as features, a probabilistic generative model is introduced to recognize American football plays in [27]. Swears and Anthony [37] propose a non-stationary kernel HMM approach which processes player…

.