Figure 1.
Plane to plane mapping which transforms points in one plane to points in the other plane using the homography matrix.
Figure 2.
Exemplary frame from a video (left) and the background calculated for this scene (right).
Figure 3.
Difference image obtained from subtraction of the background image from the current frame (left), and then single channel grayscale converted (right).
Figure 4.
Binary image after thresholding (left) and result after applying the opening filter (right).
Figure 5.
Final result after applying the foreground mask to the current frame of the video (left) and binary image showing pixels differing from the previous frame and used for calculating the movement weight for each particle (right).
Figure 6.
Screenshot of the tracking of one of the players by means of the integral histogram approach.
The grid is shown on the original frame for illustration purposes, while the tracking is actually done on the foreground mask filtered frame to suppress the background from the calculations.
Figure 7.
Horizontal candidates positions plotted over time (left) and vertical candidates positions plotted over time (right).
Matching candidates on both relationships are shown in blue, interpolated candidates in red.
Figure 8.
Candidate trajectories found by the algorithm as proposed in [20] (left) are shown in blue, with interpolated values continuing a found trajectory shown in magenta.
The Hough based algorithm results are shown in the middle and right for an exemplary video sequence with 8 ball contacts.
Table 1.
Results for the tracking of the players – Particle filter method.
Table 2.
Results for the tracking of the players – Integral histogram method.
Table 3.
Results for the tracking of the ball.
Table 4.
Results for the time efficiency of the algorithms.