Cost-Optimized Event Detection in Football Video
(2020) In Master's Theses in Mathematical Sciences FMAM05 20201Mathematics (Faculty of Engineering)
- Abstract
- In this thesis we aim to investigate if it is possible to detect events in football video using deep learning methods. The performance of a few different models using video input of varying frame rates is evaluated to find the most promising approach. We also evaluate if a proprietary dataset consisting of estimated player positions and movement can be used to detect events as it would be cheaper than processing raw video. The footage provided comes from two stationary cameras aimed at the left and right side of the pitch respectively which distinguishes our work from most previous research which is done on broadcast video where models can find events by detecting differences in zoom, replays, etc. We also provide novelty by implementing... (More)
- In this thesis we aim to investigate if it is possible to detect events in football video using deep learning methods. The performance of a few different models using video input of varying frame rates is evaluated to find the most promising approach. We also evaluate if a proprietary dataset consisting of estimated player positions and movement can be used to detect events as it would be cheaper than processing raw video. The footage provided comes from two stationary cameras aimed at the left and right side of the pitch respectively which distinguishes our work from most previous research which is done on broadcast video where models can find events by detecting differences in zoom, replays, etc. We also provide novelty by implementing our system in an online fashion to be used in real-time.
To limit the scope of the thesis we constrain the events to corners, kickoffs and free kicks which are relatively similar in their positional structure across games. The results we get are very promising. The frame-based models get decent results, while the models using the player position data achieves extraordinary results while being significantly cheaper to use. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9017573
- author
- Ericsson, Richard LU and Arpe, Joakim LU
- supervisor
- organization
- alternative title
- Kostnadseffektiv händelsedetektion i fotboll
- course
- FMAM05 20201
- year
- 2020
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Deep learning, football, video recognition, event detection, machine learning
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- LUTFMA-3416-2020
- ISSN
- 1404-6342
- other publication id
- 2020:E36
- language
- English
- id
- 9017573
- date added to LUP
- 2020-06-15 15:47:35
- date last changed
- 2020-06-15 15:47:35
@misc{9017573, abstract = {{In this thesis we aim to investigate if it is possible to detect events in football video using deep learning methods. The performance of a few different models using video input of varying frame rates is evaluated to find the most promising approach. We also evaluate if a proprietary dataset consisting of estimated player positions and movement can be used to detect events as it would be cheaper than processing raw video. The footage provided comes from two stationary cameras aimed at the left and right side of the pitch respectively which distinguishes our work from most previous research which is done on broadcast video where models can find events by detecting differences in zoom, replays, etc. We also provide novelty by implementing our system in an online fashion to be used in real-time. To limit the scope of the thesis we constrain the events to corners, kickoffs and free kicks which are relatively similar in their positional structure across games. The results we get are very promising. The frame-based models get decent results, while the models using the player position data achieves extraordinary results while being significantly cheaper to use.}}, author = {{Ericsson, Richard and Arpe, Joakim}}, issn = {{1404-6342}}, language = {{eng}}, note = {{Student Paper}}, series = {{Master's Theses in Mathematical Sciences}}, title = {{Cost-Optimized Event Detection in Football Video}}, year = {{2020}}, }