Skip to Main Content
The utility of automatic systems to visually identify vehicles based on the recognition of their license plates is nowadays unquestionable. They can be applied in very different scenarios, like access control, calculation of parking fares, automatic payment of tolls or parking fines, traffic control, etc. In the literature numerous works can be found proposing solutions to the automatic license plate recognition (LPR). Nevertheless most works propose specific LPR systems for particular applications, imposing applicability restrictions that limit their use. Restrictions range from the assumption of stationary backgrounds and fixed lighting conditions to a limited (or null) speed for the vehicles. These restrictions are sometimes overcome through expensive and complex systems. This paper proposes a versatile LPR system that yields good results running on a low-cost platform in non-controlled environments. Our method, based on multiple neural networks, is fast enough to be applicable in camera-in-motion applications and exhibits a high rate of success (up to 95%) in scenarios with non-stationary backgrounds, and varying lighting conditions.