Tennis Player and Ball Tracking

Tennis Player and Ball
Tracking
Yeshwanth Malekar
CONTENT
Motivation

Results
Challenges

Future Work
Previous Work

References
Algorithm
Detection
Solution
Problems
Motivation
 Tracking Tennis players and ball in a Tennis match
 Useful for automatic tennis video annotation.
 Player and ball trajectories contain high level
information which can be used for training purposes and
designing strategies.
Project Goal
Challenges
 Video taken is not from a steady camera.
 The speed and size of the ball pose real problems.
 Most of the times, ball undergoes either occlusion or gets
blurred in the background.
 Chances of false detection are high.
Previous Work
 Most of the previous work is done on a video taken from
a steady camera.
 They also used Kalman filter and Particle filters to predict
the location of foreground objects.
 These filters need a steady camera to work on and a
model of the movement of players and balls.
Algorithm
1. Extract Background image.
2. Frame difference method.
3. Morphological Operations.
4. Blob analysis.
5. Detect players and ball.
 Background extraction using median function.
Bad background examples:
Too few Frames:
Over all frames:
 Foreground Separation
Player and Ball detection
Morphological Operations
Conditions
 Added extra boundary conditions for the ball.
 Finally centroid conditions to keep track of the players
and ball in the following frames.
What if camera moves?
The whole court is detected in the
difference image instead of just
Foreground.
Simple solution
 Skipped through the frames with more than required blobs.
 Continue detecting the players when the video is stabilized.
 Saved the centroid of the players detected in all the frames.
Adjustments:
 Added the difference of the centroid positions of last two images to the
previous image.
 We have new approximated candidate positions for the blank frame.
Advantages of this solution
 It is very simple technique to implement.
 Players detection is excellent.
 Even if the ball is occluded or blurred, it is detected.
Advantages (Cont.)
Problems encountered
 The ball moves at high speeds and changes its direction
rapidly.
 If detection in a frame is missed, keeping track of the
ball in the next frame with this approximation method is
not possible.
 False detections influence further false detections.
Problems (Cont.)
Results
Results (Cont.)
Result (Doubles)
Future Work
 Better ball detection techniques for a moving camera.
 Would also like to work on different ball games such as
soccer.
References
 Hira Fatima et al, “Object Recognition, Tracking and Trajectory Generation
in Real-Time Video Sequence”, International Journal of Information and
Electronics Engineering vol.3, no.6, pp. 639-642, 2013.
 Amor Salorpour et al, “Vehicle Tracking Using Kalman Filter and Features”,
Signal & Image Processing, an International Journal (SIPIJ), Vol.2, No.2,
June 2011.
 F. Yan and W. Christmas and J. Kittler, “Tennis Ball Tracking Algorithm for
Automatic Annotation of Tennis Match”, BMVC, pp 619-628, 2005.
Questions