The cooperation between Kapsch TrafficCom Inc. – an international supplier of intelligent transportation systems – and the image processing experts of the Austrian Institute of Technology (AIT) provides improvements in detection and readability of automated license plate recognition systems.
Enhancement of existing character recognition algorithms
Kapsch TrafficCom Inc. uses an automatic character recognition system within their tolling applications. A core component of this system is a Convolutional Neural Network (CNN) for single character classification.
Fig. 2: net structure of CNN for character classification
The CNN was enhanced by AIT to the latest state of the art in order to gain better classification results.
Impact and effects
Kapsch TrafficCom Inc. is supplier of electronic tolling- and traffic management Systems
A vital part of these systems are components which detect and recognize license plate numbers of camera images. The focus herein is in particular the maximum read rate of these license plates.
By means of complex mathematical algorithms it is possible to improve image quality and therefore to improve license plate read rate.
This allows using cheaper camera hardware with lower image quality and still to obtain high license plate recognition rates. Another application is to use a single camera not only for one lane but to use it for two or even more lanes. This will lead to more cost efficient applications within a traffic management system.
Abb. 3: automatic license plate recognition
Altogether, the improvement of the license plate recognition system leads to higher overall performance and to less need for human post processing.