Sunteți pe pagina 1din 1

Computer Vision Assignment - License Plate Recognition

TI2716-B

Introduction Week 1 Segmentation


With this poster we would like to inform you about the progress we have made In order to recognize various license plates, we have applied the following
during the first sprint. Our aim is to make an application that recognizes license 2 steps:
plates. To do this it is necessary to only focus on the license plate in an image. 1) Use a threshold to seperate the license plate from the background.
Subsequently, you can start performing character recognition on the plate. Then 2) Crop the original image to only show the license plate.
you can return the string of the license plate with the associated frame number 3) Use a threshold to seperate the characters from the background.
and timestamp. Our focus of this week was to recognize Dutch license plates 3 4) Use the bpropagation function to remove all edge objects in the
on non-moving cars and to label all the characters on the license plates. binary image.
5) Label the characters in the license plate using the label function.
1 4 So to seperate the license plate from the background we used multiple
thresholds. First we took three colour samples from the video containing the
yellow color of the license plate. Then we created three scatter plots to com-

GUI 5 pare the three colour channels (RGB) of the video frame. The result is shown
in figure 2.

To show the results of our license plate recognition application we have build
a GUI containing the following aspects. Our GUI, shown in figure 1, consists of
an axis graph that shows the original video of the cars, an axis to show the fil-
tered license plate of the showed frame, a table to store the results and buttons
to start/stop/restart the video. The table will contain a string of the characters
To do
found in the license plate, the frame number and its timestamp. The style we
have chosen for the GUI is traffic lights. This explains the colors green, red and Performing the steps given in the segmentation
orange for the buttons. We decided on this theme, because it is applicable for section currently gives a decent result. Most li-
our application. cense plates in the categories 1 and 2 return
Figure 2: Three scatter plots comparing red-blue (top-left), red-green (top-right)
Currently our license plate recognition program cannot yet turn the characters an image similar to image 5 above. and blue-green (bottom-left)

into text. However it does not work on every car. In


two cases we have found that one character is After applying the thresholds, we used various binary functions to select the li-
attached to an edge object. It then gets re- cense plate and remove all other objects in the binary image (image 1).
moved during step 4 of the segmentation. Next we used the imcrop function provided by Matlab to cut out the license
plate (image 2). In order to recognize the characters in the license plate we
made use of another threshold. This time we used the threshold calculation
algorithm provided by the slides from the course. We applied this to the green
Figure 3: License plate that removes a character
colour channel as this seems to give the best result (image 3).
It also does not yet work on frames that con- Next we needed to remove all other objects from the binary image. We did
tain 2 license plates. Our code currently crops this by subtracting the output-image of the bpropagation function from
a frame when it detects a license plate. It does image 3. This operation removes all edge objects from the binary image
not yet crop it twice for each license plate in (image 4).
the frame. Finally we labelled all characters by using the label function (image 5).
Foreign license plates are also not recog-
nized yet. Currently we look at yellow license For the next sprint we will focus on recognizing the different characters and
plates only. store them in a String.
Figure 1: The design of the GUI

Zo van Steijn & Emma Jimmink - Group 9

S-ar putea să vă placă și