Cryptography Reference
In-Depth Information
(a)
(b.1)
(b.2)
(b.3)
Fig. 9.11.
(a) Original video clip. (b.1)(b.3) The representative motion flows se-
lected with threshold ǫ = 0, 0.05, and 0.1.
storage space required, we remove redundant points from a trajectory, leaving
only a few necessary points to represent it. We then use the Douglas-Peucker
algorithm introduced in Section 9.2.3 to select the necessary control points
from the trajectory. The algorithm starts by using a straight line which is
called the anchor line to connect the start and end points of the trajectory.
Once the perpendicular distance between any intermediate point and the an-
chor line is larger than a given threshold, the trajectory is split into two
segments via the farthest intermediate point. The process continues until all
the perpendicular distances are smaller than a pre-set threshold. Finally, the
chosen intermediate points and the two end points are reserved as the control
points of the trajectory.
We use six positive real numbers (x+,x−,y+,y−,d,t) to represent a con-
trol point on a trajectory, where d denotes the cumulative length of the tra-
jectory from the first control point to the current control point. Here +/−
denotes the cumulative positive/negative movement along the x−or y−axis
from the first control point to the current control point. Now, let Q and D
be the trajectories of the query and a model in the database, respectively. In
a QBS case, we normalize the length of both trajectories into a unit length
before comparing them. This guarantees the requirement of scale invariance.
Therefore, the parameters d, x+, x−, y+andy−of each control point on the
two trajectories have to be normalized by dividing them by the length of Q
and D, respectively.
According to the six parameters of a control point, d and t are utilized to
make a fair comparison. We align both Q and D by calculating the length d
