Indexing, Object Segmentation, and Event Detection in News and Sports Videos - Multimedia Database Retrieval: Technology and Applications - page 202

2. Run Plays : During a run play, first the motion is lateral as the runner gets the
ball. Then the camera zooms in, to track the movements of the ball carrier. This
zoom in provides the perception of high intensity motion. At the end, the camera
laterally tracks the movements of the ball carrier. Therefore, the movements of
a run play involve firstly low intensity lateral movement, followed by short high
intensity lateral movement and in the cud low intensity lateral movement.
3. Kickoff/Punt (K/P) : This is the first category of kicking play. The kicker starts
by kicking the ball high into the air. This motion causes the camera to rapidly
zoom out to capture the kicked ball. After the kick, the camera zooms into
the player who has the ball, and tracks the movements of the ball carrier.
Therefore this play has movements that involve firstly the high intensity motion
of zooming out and zooming in with horizontal direction movement, followed by
low intensity lateral motions.
4. Field goal/Extra point (FG/XP) : This is the second category of kicking play.
The ball is long snapped (short underhand throw) to a holder who sets the ball
up to be kicked by a kicker. The majority of the movement is low intensity with
most of it coming after the kick, when the camera is tracking the kicked ball as it
sails towards the goal post. Therefore the majority of motion in this category is
vertical and low intensity.

The global motion of camera, the intensity of motion and the direction of
motion provide valuable discriminating information regarding the different types
of plays. In the current work, the motion based feature set is developed by
utilizing the intensity of the motion descriptor and dominant direction descriptor in
MPEG-7. The motivation behind using a combination of the two descriptors comes
from analyzing the different plays as defined above.
The magnitude of motion vectors was calculated by extracting the encoded
motion vector given by coordinates

from the macro blocks within P frames
of the MPEG-1 video stream. The magnitude and the direction of the motion vector
are given by the following equations:

( x , y )

x 2

MAG mv = + y 2 (7.44)

y
x )

ʘ mv = arctan ( (7.45)

According to the MPEG-7 description [ 218 ], the standard deviation of the
magnitude of the motion vectors can be used to derive the intensity of the motion
descriptor. The descriptor takes on the value of 1-5, with a low value meaning
a low intensity of motion. Experiments done by using five levels showed that
most of the motion descriptors were quantized into two or three levels. Thus to
provide better motion activity resolution, the descriptor was quantized into 12 levels.
Similarly, according to the MEPG-7 description, the dominant direction descriptor
is calculated by quantizing the angles of the motion vectors into 8 levels. In this
work, the same eight quantization levels were used to define the dominant direction
descriptor.