Automatic Number Plate Recognition (ANPR) seems to serve little purpose to the man on the street, but for authorities, it has become an invaluable tool.
Optical character recognition (OCR), the technique used to read license plates from an image, came about through the invention of two devices from separate sources at roughly the same time. First, in 1912, Emanuel Goldberg, a scientist who contributed to almost all aspects of imagining technology in the first half of the twentieth century, patented a machine that read characters and converted them into telegraph code and then transmitted the messages without human intervention. Secondly, in 1914, the optophone was invented. The optophone "scanned text and generated time-varying chords of tones to identify letters." But although it was a useful proof of concept, it was impractical for daily use as it could only read one word per minute but combined, these ideas would lead to the development as we know it today.
It would be nearly 40 years before OCR was to be used for the printed page and since the 1990's, when computers became more widespread, OCR is common place and achieves a reliability of 99% on clearly printed text. With all the varieties of text and fonts available this is quite a feat, but on an average page of 1000 words, leaves 30 words unrecognized. These statistics aren't acceptable for areas where 100% accuracy is imperative.
ANPR systems have several more challenges than scanned or photographed images contend with. There are varying quality cameras and the object can be too far away, therefore providing low image resolution; motion blur from fast moving cars; poor lighting and contract caused by shadows; over exposure and reflection; objects such as dirt and tow bars obscuring the plate and circumvention techniques.
APNR systems use a combination of three key components: the camera, the software, and the hardware it runs on.
The cameras vary depending on the country but include a combination of CCTV, mobile camera, and infrared. Infrared cameras have the advantage of being able to take images whatever the lighting conditions. To cope with varying car speeds, the shutter speed of the camera is typically set ta 1/1000 of a second which can capture images without blurring as cars move at speeds of up to 120 mph (190 km/h).
The software uses seven key algorithms to recognize the plate: plate localization finds and isolates the plate from the rest of the image; orientation and sizing compensates for the skew of the plate and adjusts it to correct the size; normalization adjusts the brightness and contrast; character segmentation which finds the individual characters; OCR which recognizes the characters; syntactical/geometrical analysis which checks the characters and positions according to the country's rules; finally there is the averaging of the recognized values which takes the images form the cameras (systems usually employ more than one camera per police car or static system) and produces a more accurate result. All of this is carried out in less than 250ms.
The hardware no longer needs to be specialized and runs on standard off-the-shelf components which can be stored locally in the police car or wherever else the cameras are. If human intervention is required to make out the details of the plate, the image can be sent to the central location where operatives can decipher the details.
We may not like license plate recognition and the invariable speeding fines that we may receive, but beyond that that are responsible for tracking stolen vehicles and can help in the fight against crime.