Authors:
Giorgio Grasso
and
Giuseppe Santagati
Affiliation:
Facoltà di Scienze, Università degli Studi di Messina, Italy
Abstract:
The recognition of car license plates has a variety of applications ranging from surveillance, to access and traffic control, to law enforcement. Today a number of algorithms have been developed to extract car license plate numbers from imaging data. In general there two class of systems, one operating on triggered high speed cameras, employed in speed limit enforcement, and one based on video cameras mainly used in various surveillance systems (car-park access, gate monitoring, etc). A complete automatic plate recognition system, consists of two main processing phases: the extraction of the plate region from the full image; optical character recognition (OCR) to identify the license plate number. This paper focuses on dynamic multi-method image analysis for the extraction of car license plate regions, from live video streams. Three algorithms have been deviced, implemented and tested on city roads, to automatically extract sub-images containing car plates only. The first criterion i
s based on the ratio between the height and width of the plate, which has, for each type of plate, a standard value; the second criterion is based on the eccentricity of the image on the two dimensions, i.e. the projection histogram of plate number pixels onto the reference axes of the image; the third criterion is based on the intensity histogram of the image. For each criterion a likelihood is defined, which reaches its maximum when the tested sub-image is close to the standard value for the type of plate considered. The tuning of the methods has been carried on several video streams taken during travel on busy city roads. The results for the overall recognition rate on single frames is around 65%, whereas the multi-frame recognition rate is around 85%. The significant value for the performance of the method is the latter, as typically a license plate is visible in 5-10 frames. Based on three parameters ranking, the same system can potentially distinguish and identify a wide range of license plate types.
(More)