Automatic Car Plate Reading at UPV

	Automatic Car Plate Reading
	Image and Video Processing Group of the Departamento de Comunicaciones Universidad Politécnica de Valencia. SPAIN

The image and video processing group at the Communications Department of Technical University of Valencia (Spain) has continued working in automatic car plate reading after VISOR BASE projetc, supported by the European Union.

Our work continues now funded by the Spanish Ministry of Science and Technology within a CICYT project.

Our technoloy is now mature enough to be integrated into products for different applications: parkings, motorways, city parked cars monitoring, etc.

Contacts with the company STN have started to create products based on Intel-Linux architecture. Appart from products, we are also developping solutions for specific problems that require automatic car plate reading.
We are NOT dealers or resellers of other parties' products. All the technology has been developped within our research group, from image aquisistion to results management.

Some aspects of our technology are:

We use standar video (PAL) as input. This implies low cost of imaging hardware.
No external trigger is needed in order to signal the presence of a car in the field of view (although if it is available it can be used).
Currently we are able to process 25 fps (full rate video) on a Pentium IV (@ 50% CPU-load).
The cars to be read can be static or moving. The system is able to read plates at 120 km/h (speed limit in Spain).
The system can be static or on board vehicles.
No special illumination is required in daylight conditions.
During night or indoors infrared or visible illumination can be used.
Due to limitations in standar video resolution, the car must occupy at least half of the width of the video image in order to read reliably the plate. This is approximately the width of a lane.
The system DOES NOT GUARANTEE 100% reading efficiency. In fact no system can guarantee that (not even a human reader !!). So, the possible applications should be restricted to those where a (low) percentage of erroneous or impossible readings might occur.
Since using 25 fps yields normally many images of the same car are available, all the readings from the images of the same car are integrated in order to give a more reliable result.
Performace of car plate readers is difficult to compare. In other words, claiming a success of 99% says nothing if you do not say how many cars make the testbench, under wich resolution and illumination conditions, which background, etc. One of the goals of our research is to try to build a DATABASE which can be used as a reference to compare different systems

Some examples of the kind of situations that our technoloy can deal with are shown below.


Click the picture for MPEG video (14.5 MB)	Click the picture for MPEG video (2.5 MB)	Click the picture for MPEG video (11 MB)

If you need further information do not hesitate to Contact us .

January 9th, 2004