Vehicle License Plate Detector

[z3n]

Hie. I am currently trying to locate the license plate of a vehicle in any image. By locating this feature i would be able to progress further to read the characters on the license plate. I have succeeded in one method which uses the Color Location VI. However this could be developed further into a more robust system as the success rate is about 10 out of 18 images. This is due to the different environments and reflection of the plate.I have read through some conference papers in which they suggested to use edge detection technique whereby the two parallel lines of the plates will be detected. I am eager to know whether this could be done in LabVIEW or are there any other possible techniques to locate the position of the license plate so that it could be translated into the ROI.

[DJ_L.]

Hey z3n,

You will be able to use the edge detection technique in LabVIEW to detect parallel lines, such as the ones on the edges of a license plate. I am assuming that your images will be different for different cars with multiple styles of license plates. Therefore, you might have to use, not only the edge detection method, but even some other techniques that come with the NI Vision software that you can get with the Vision Development Module. Some other possible techniques that you can try are the Pattern Matching functionality, the Shape Matching, and even some of the Color Matching or Color Pattern Matching functions. All of these functions are accessible in LabVIEW by having the NI Vision Software. You can learn more about the Vision Development Module by going to www.ni.com/vision. Also, you can request an Evaluation Version of the Vision software to try out for 30 days. You can request this online at the NI Vision Development Module page.

I hope this answers your question. Let me know if you have any questions. Thanks, and have a great day.

Regards,

DJ L.

[z3n]

Thanks for the reply. I am still not quite sure how could the detection of parallel lines using edge detection techniques could be done in LabVIEW. However there are many methods to detect the position of the plate. I am still working on some of it using thresholding, particle analysis and then color location to locate the license plate. However, i have noticed that light variation is a major contributor to the success or failure of the location of the plate. To have a more robust color location, i have read a book which recommends the use of the HSL space. As the application is done using a WebCam, the pictures obtained are in the RGB24 format. Then by using Image Cast, i convert the image to the HSL space and then saving this image under the AIPD format. However, using this AIPD picture file, i couldn't create a template for the color location in Vision Assistant. I would like to know whether this is the proper way to convert RGB to HSL space and whether it is possible to create templates using the HSL pictures. This is quite crucial to me as learning templates from the HSL space would enhance the success rate of the location of the license plate. Thanks for your help in advance

[DJ_L.]

Hey z3n,

There is an easier way to convert your RGB images to the HSL space. There is a step in the Vision Assistant called Extract Color Planes, and you can chose to extract the HSL - Luminance plane which will give you your HSL (8-bit) image. Also, this same function is available in LabVIEW. The VI you can chose in LabVIEW is called IMAQ ExtractColorPlanes.VI and you can set the Color Mode to HSL. You can create a template from the resulting image if you would like. 

I hope this helps. Let me know if you have any questions. Thanks and have a great day.

Regards,

DJ L.

[z3n]

Thanks. I guess i know how to play around with the HSL plane. I have already created an algorithm which could work quite well on some sample images i have taken using my digicam. It is a combination of thresholding, edge detection and as well as clamping. However, i have realized that thresholding is one of the most important procedure in the algorithm. Before thresholding, i equalize the image. This work well for cars which are not dark in colour (dark green, black and etc). For other colours like silver, red, white, it works pretty well. The current threshold i am using is the 'clustering' type. However, this is very prone to the lighting conditions of the surrounding environment. Are there any methods available which will be able to analyse the picture automatically so that it would provide the best thresholding range for my application or are there any kind of image enhancements i could make to the original image so that lighting would be a major factor? Anyway, the license plate here has a black plate with white characters on it. I think it is one of the reasons why a dark colored car is harder to threshold with the current range.Thanks for your help!

[Nurvasi]

Hi,

I am newer with Labview. I start to use it to implements a program to can read the License plates. I read you are doing the same, please can you help me to start?? I think labview is very dificult....

If you can help?, we can help to each other in the future...  if can send your beginers programs...

Please help me, I don´t know how can I start...

[DJ_L.]

Hey Nurvasi,

You might also want to refer to the following discussion forum matching colors from a saved template.

DJ L.

[zio]

Hi,
please don't be offended, there is nothing better to learn something than to do it yourself.
But just in case you need the solution, there are some great guys out there who have been doing this for about ten years.
http://www.neuricam.it/main/solutionslist.asp?i2=Transport_and_Security
When I last read about them,  their chips were based on neural algorithms, which work fine with specific problems, but don't tell you much about exactly how they get to the answer.
cheers,
S.

[Oliveirakol]

Dear Friend,

I am just doing it now in Brazil. I need to read Licence plates from cars. I am using LabVIEW Vision software. I can read the licence plate if the car is stopped. If it is moving above 5 km/h I am having trouble. One important thing is to align the image to make the software read the plate as if it is in front of the vehicle. Is important to use a rignt camera with have WDR, BLC compensation. I am using the black and white righ resolution graphic card.

you can contact me by mail oliveira@kolbr.com.br  OK?

[muks]

Why would you dig an old thread like that. Choose a camera having a good shutter speed. Use co-ordinate system for alignment.