Automatic number plate recognition (ANPR) is a great addition to standard traffic video processing in GoodVision Video Insights, but there are several issues with ANPR camera setup that impedes its accuracy. In this article, we share the tips on how to improve your camera setup to increase the ANPR accuracy.

How can I get ANPR?

This is not rocket science. 🚀

It is easy to blame the ANPR engine when your ANPR accuracy rate is low. However, when you start considering an ANPR processing for your footage, the most important question is:

CAN YOU READ THE LICENSE PLATE YOURSELF?

🚫 NO:

If your answer is NO, there is no way any software can do it. Artificial intelligence doesn't have deductive skills as the human counter has. Therefore, if your thinking is that the license plate is not readable right away, but if you look into it you try to guess the shapes of numbers and letters according to some blurred out object, this is a NO-NO. Artificial Intelligence cannot do this. 🤖

With this argument, we already eliminated loads of video types and determined what you actually can use. In GoodVision Video Insights, the ANPR Video Processing is suitable for STANDARD Processing only because only the scenes for this type of processing can satisfy the requirement on the size of the license plate - read below. Another NO GO is blurred, fogged, undersaturated, or under/overexposed footage.

If you want to find out more, how these aspects can affect the overall accuracy not just in the ANPR processing check out our webinar explaining this topic in more detail.

✅ YES:

If you can read the licence plate by yourself, let's check more detailed requirements.

Video resolution and FPS:

It is logical, that with a higher resolution of your camera and with higher FPS the quality, readability, and accuracy of your video processing boosts. So if you can, go for as high as 4K. For example, 8 MPix cameras are needed for highway or street monitoring.

However, we are aware, that not everybody has the option to record in 4K resolution so the general rule of thumb is that the license-plate width shall be at least 100 pixels in width. The lower the resolution, the bigger portion of the image the vehicle must take. But beware to zoom too close, otherwise the video will not be suitable for standard processing.

Camera frame rate (FPS) is a very important parameter that limits the maximum speed of vehicles for reliable license plate recognition. Please, see the table below that will help you choose the minimal FPS for a specific vehicle speed range.

Camera Angle:

The only thing to keep in mind regarding the camera angle is that it should not exceed 45° in the vertical or horizontal direction - see the graphics below. Anything above 45° is complicating the detection.

Your camera shall have direct visibility of the vehicles and their license plates at the point of the scene, where you need to recognize it, so try to avoid the camera positioning where the vehicles overlap each other, like a direct angle.

NOTE: Don't forget that you still need to meet Standard Processing requirements, scene wise. If you are not sure, check the article below.

And here is how it should and shouldn't look like in practice:

NOTE: Both of these scenes may look suitable for ANPR recognition, however, not both of them are good for Standard Processing. Why? Because the vehicles are overlapping significantly on the first image and you will not obtain proper trajectories. When Uploading footage for ANPR recognition it has to be suitable for the Standard processing as well.

NOTE: In the first scene the camera is placed too low which may cause overlapping of the vehicles, therefore, block the view of the vehicles behind which trajectories won't be captured correctly and the licence plate number not properly assigned.

We hope that this guide will simplify your choice of footage or the camera used for this purpose.

If you are still not sure about the usability of your footage contact us!

Did this answer your question?