Quantitative Evaluation of the Estimation Accuracy of Object Displacement and Velocity in Soccer Games

In this paper, we estimated the displacement and the velocity of a 3D object using perspective displacement field model and quantitatively evaluated the estimation accuracy in soccer games. To find the coordinate transformation between two successive frames due to the camera motion, e.g., panning and zooming, we computed motion vectors by block matching algorithm, and selected dominant motion vectors only from the histogram. Then we substituted the selected motion vectors into the planar motion model equation and solved it in the least mean square sense. To estimate the displacement and velocity of an object, we found the coordinate transformation between a soccer image and the soccer field model by using perspective displacement field model. We applied the technique to synthetic and real soccer image sequences. Experimental results show that the estimated displacement and velocity have error on average 4.7% and 4.5% respectively.