There are two main steps to extract the key frames.
First, the parameter qualified the character in every
frame should be figured out. Second, we should check
226
whether the parameter is the key character. In the lane
surveillance system, the useful information includes
the vehicle-license-plate, the model, and the color of
the vehicle and the portrait of passenger in the vehicle.
We call all the information mentioned above as valid
information in the lane surveillance system. We don’t
care about the other background information. So we
use the valid information in every frame as the key
character to be qualified, according to these qualified
key characters, we can get a peak value in the curve
which is used to describe the change of these
characters according to the time axis. Then we select
the frame at the peak value as the key frame.
From analyzing every frame in lane surveillance
system, the vehicle information must include the
vehicle-license-plate; what is more, the vehiclelicense-plate
lies in the downside of vehicle. So the
nearer the vehicle-license-plate is to the bottom of the
image, the more information of the vehicle can be
captured, that is to say, more body of the vehicle and
more information about driver can be included in the
frame. If the vehicle-license-plate of the vehicle can be
located, we use the distance between the location of
vehicle-license-plate and the bottom of the image as
the eigenfunction. The less eigenfunction, the more
information of vehicle can be captured in the frame.
Presently, there has been much research done on the
vehicle-license-plate locating. Among these main
methods, are methods based on color information,
combining wavelet transform with morphology,
projection of binarization, and energy filter [5,6,7,8],
etc. In order to extract the key frame accurately and
quickly, we select the method based on adaptive
energy filter to locate the vehicle-license-plate. In this
selected method, because that the energy of the vehicle
license-plate is highly concentrated in horizontal
direction, the candidates of vehicle license-plate can be
fast and roughly segmented by an adaptive high-pass
energy filter , the candidates are validated and
modified according to the fine texture feature of
vehicle license-plate.Thus,the vehicle license-plate
can be accurately located . After we obtain the
location of the vehicle-license-plate in every frame, we
select the frame in which the location of the vehiclelicense-plate
is nearest to the bottom as the key frame.
Experimental result of is shown as follows.