Schmitz GmbH, Regensburg - the specialists for digital character recognition

Character recognition

Vehicle Access Control by means of Automatic Number Plate Recognition (ANPR) is based on methods of image processing. The digital or digitised images of a video camera are evaluated by means of complex algorithms according to sign information to be able to automate the processing of the data won for different applications.


In a first step a picture of the vehicle must be provided for the computer. This picture is given according to the used resolution, for example, in a format of 800x600 pixels. Task of the software is then to reduce this amount of data (800x600=480.000 bytes for s/w) to a few bytes (7 bytes for the example plate R-EK 275).

NPR with CAR-READER Image 1

The camera "sees" a certain gray-scaled value or colour value for every pixel. Applying all these values on a surface an almost arbitrary "mountainous region" arises, in which the plate is not readable as used but hidden according to the geometry of the appearance of the characters.

to the top

The following three tasks are to be solved software-technically:

to the top


Triggering is the mechanism capturing images. Good image processing systems deliver pictures on a real-time basis, i.e. 25 pictures per second. This means each 40 milliseconds a picture is available. The better the pictures are dissolved, i.e. the more pixel per picture exist, the greater the requirement for the image processing system to guarantee real time.

If you want additionally evaluate the frames by OCR to get plate information, the workload of the system rises further. If you need to, for example, 40 milliseconds for the OCR, the frame rate is divided in half.

You can avoid this workload by avoiding permanent image analysis and waiting for an external signal giving the time to start recognition.

Technical solutions for that are for example inductive ground loops, light barriers or impulses of third party systems.

An other possibility is to begin the character recognition by internal or external motion detection.

Which method is best suited depends on the respective application.

to the top

Recognition rate

We do not promise 100% of recognition rate.

It is better to quote a customer, who is security service manager at a huge company close to Regensburg: he claims that the recognition rate is better than 98 % With more than 1000 entries and exits per days that means that only approximately 20 problem cases arises.

Some problems can be handled by configuring the system with “one character tolerance”. The rest is done comfortably by correcting the platestring manually.

Because the efficiency of the recognition algorithm is dependent on the quality of the delivered image, the installation and configuration of the system plays an important role. Here in particular the choice and position of the camera, the optics, the catch range of the lane are to be considered.

We therefore offer workshops or recommend to let install the systems by trained partners.

However, our most important concern, as the developers of the system, is the continual improvement of the recognition and the capabilities of the software features.

to the top