Football-Player-Tracking
Computer vision based and deep learning based framework for player tracking and analysis in football videos
Description
In this project i aim is to do Computer Vision based analysis on a live football video stream or a recorded football match which can be used as a football analytics tool. My plan is primarily based on the ideas preseneted in this paper [1] with the end goal that the same system can be portable to be used in other sports too. Generally a video stream from a live match only gives view of a certain section of the pitch(shown in below gif), one of things i aim to is to be able to convert it to whole pitch representation i.e a panoramic view for easier interpration in terms of a global view.
Along the with the panoramic view generator, i will also try to make 2d pitch representation of the pitch to map the live player positions on to it.The 2-d mappings will be used to calculate various foobtall metrics like player trajectories, team structures etc.
I plan to read to more research papers on basis of the use-cases of which tools will be required to do the above mentioned stuff
Some of the topics covered in this project would be:
Position mapping - positions from the video are mapped to a physical reference frame, in our case the soccer pitch
Detect and track objects of interest - In case of team sport video analysis, these are typically the
teams’ players and the ball possibly using deep learning object tracking algorithms
Datasets that can be used:Soccer video & Position dataset
Aims
- Using homography to
1.1 Create panoramic views
1.2 Projecting current video frame to panoramic view
1.3 Projecting current video frame to to 2d pitch representation - Try out various player tracking approaches to identify player positions on the 2d homagraphic pitch view using CNN’s and other such architectures
- Calculate various football analysis metric possible,like
4.1 Player trajectories
4.2 Team structure