What Algorithms Are Used to Let Automatic License Plate Recognition Technology Work?

Asenqua Tech is reader-supported. When you buy through links on our site, we may earn an affiliate commission.

Licence Plate Recognition Technology (LPR) or Automatic Number Plate Recognition (ANPR) is software that reads a register plate for example, but has no sound, by transforming images into characters, for the purpose of traffic control and law enforcement. However, when working backwards, these systems can be used in a large number of industries in many ways.

But How Does it Work?

Afterwards, an image-processing algorithm locates the plate, and then normalises and cleans up the image, so that the image-processing program is able to ‘read’ the alphanumeric characters that make up the plate. Some systems are single-lane capable, but able to be run in real-time (around ¼ of a second, or 250 milliseconds) to extract the plate information from an image. Some other such programs are multi-lane capable, but take a little bit longer to extract the same data.

Essential Algorithms

Given that plates from different regions and countries display fonts, sizes and positions of letters and numbers in varied ways, LPR hardware and software must make 7 different algorithms work with a great many variables from plate to plate:

Plate Localisation: Identifies a plate in a photo and isolates it, shedding distracting information so the LPR software can concentrate on what it’s looking for: the character strings.

Orientation and Sizing: to normalise the image and make it easier for the reader, the algorithm compensates for any skew of the image and scales the dimensions to a uniform size.

Normalisation: The brightness and contrast are adjusted according to a plate’s information.

Character segmentation: Equidistance and colour segmentation.

Optical Character Recognition (OCR): Another chance for the text to be messed up; this is where the separate characters get translated into an alpha-numeric entry.

Syntactical/Geometrical Analysis: The algorithm compares the character and the location of each block in the image of a license plate to validate the characters using the licence plate’s rule set.

Averaging Recognised Values: The average of each algorithm and red, green, blue, brightness and contrast configurations are compared to confirm and average character values for the most optimal result we could get.

Summary

LPR is not a stand-alone piece of software; a picture is only as good as the camera that produced it.  While there is always an issue with older cameras that are not as high resolution as the modern ones, new tools designed to work with low-quality cameras are ready to help LPR-software in creating a good image and its reading.

Moreover, the increased specialisation of LPR hardware, see this reference, has helped to reduce the number of bad images by using IP cameras equipped with infra-red (IR) lighting, faster shutter speeds and better resolution setting.

Using a multitude of algorithms and the specs of both hardware and in-software, license plates are identified and, by converting images to codes, read – systems built primarily for traffic and police but that has since expanded to a daunting number of sectors for an equally daunting number of purposes.