skip to content

High-Motion Table Tennis Ball Tracking for Umpiring Applications

Result Analysis
To evaluate the performance of the new table tennis ball detection system, a set of five test sequences including a mixture of resolutions, and varying object motions were chosen to evaluate all the assorted detection challenges. Sequence 1 was extracted from the widely adopted SIF Table Tennis sequence. The service segment comprises 79 frames and contains a number of detection challenges including camera (global) and object (local) motion, a blurred object of interest (OOI) due to the lower spatial and temporal resolutions and some object occlusion. Sequences 2 to 5 were captured from actual match environments, though only the service elements were analysed. The full characteristics and detection challenges of each sequence are summarised in Table 1.


Table 1: Full characteristics and detection challenges of the five test sequences
Sequence 1Sequence 2Sequence 3Sequence 4Sequence 5
Table TennisITTF Timo BollLocal leagueHigh Throw
No of frames7945306739
Frame (pixels)352 x 240352 x 240640 x 480720 x 576512 x 384
Capture rate30 fps30 fps30 fps100 fps30 fps
Ball size area (pixels)50-17730-64 254-57224-9325-50
Ball ColourWhiteOrangeWhiteWhiteOrange
Frame size SmallSmallLargeLargeMedium
Variation in Ball sizeLargeMediumMediumLargeMedium
Object motionHighHighHighLowHigh
Global (camera) motionHighNoneLowNoneNone
Blurring of OOI41 frames7 frames28 framesNone17 frames
MergingNoNoNo2 frames9 frames
Occlusion12 framesNoNo1 frame17 frames
LightingGoodAcceptableGoodDarkDark
SourceResearch community ITTF Umpires & Referees YouTubeAuthorAuthor
To ensure an equitable comparison, key system constants were empirically determined during initialisation and maintained throughout all the experiments, with the only user inputs required being the diameter, colour and location of the ball in frame #1 of a sequence. Numerical results showing the detection rates and computational times are presented in Table 2.

Table 2: Numerical test results with the system parameters being: j=3, m=2j, k=1.3, u=v=10, g=(100|Ab – Ao|)/Ab and with the maximum number of TPT tuning iterations being set to 20

Sequence 1Sequence 2Sequence 3Sequence 4Sequence 5
No. of correct detections6442286614
No. of incorrect detections10000
No. of undetected
(frames with OOI occlusion excluded)
14 (2)321 (0)25 (8)
Detection rate (frames with OOI occlusion excluded)81% (96%)93%93%99% (100%)36% (64%)
Time for ball rise (secs) 0.56
(7 frames)
0.55
(5 frames)
0.82
(8 frames)
1.08
(9 frames)
3.52
(12 frames)

Table 1 Test sequences


Video Outputs:

These video sequences show the output of the ball detecting and tracking system, which is designed to evaluate table tennis service. The top pane shows the frame that is being processed. The yellow square box indicates the region of interest (ROI), which is the area the algorithm predicts where the ball is. If a ball is found, it will be marked by a red circle and its centre will be marked by a cross. The bottom three panes show the processing details. The left pane shows an enlarged binarised version of the ROI, where ball candidates are indicated by white objects. The middle pane shows the ball candidates that are being processed. The detection algorithm evaluates each ball candidate by checking its roundness, rounded upper contour (red) and size etc. The blue circle and cross are the predicted ball contour and centre respectively. Finally the right panes shows a plot of the x and y location of the ball versus the frame number. The plot can be used to measure the ball rise of the service.

The following embedded videos (YouTube) are for demonstration purposes only. The high quality version of these videos can be obtained on request.


The SIF Table Tennis Sequence


Output of the Table Tennis Sequence can be can be obtained on request.
The video sequence is provided for research purpose only. If you use it, please reference the paper shown at the bottom of this page.




ITTF


Output of the ITTF Sequence can be can be obtained on request.
The video sequence is provided for research purpose only. If you use it, please reference the paper shown at the bottom of this page.




Timo Boll


Output of the Timo Boll Sequence can be can be obtained on request.
The video sequence is provided for research purpose only. If you use it, please reference the paper shown at the bottom of this page.




Local league


Output of the Local league Sequence can be can be obtained on request.
The video sequence is provided for research purpose only. If you use it, please reference the paper shown at the bottom of this page.


High Throw


Output of the High Throw Sequence can be can be obtained on request.
The video sequence is provided for research purpose only. If you use it, please reference the paper shown at the bottom of this page.


Results updated in July 2011



RELATED PUBLICATIONS

More discussions of this work can be found from the paper below:

Wong, K.C.P. and Dooley, L.S., "High-Motion Table Tennis Ball Tracking for Umpiring Applications", The 10th IEEE International Conference on Signal Processing, Beijing, 24-28 October 2010.

Other related publications of this work can be found at Open Research Online.