Access control system with neuro-fuzzy supervision

Abstract

In this paper we describe a plate-recognition system for access control to restricted areas. The system we propose is based on vision, neural networks and a neuro-fuzzy system. Vision is used to detect the license-plate and to single out the characters it contains, while a neural network-based classifier is used to "read" the plate. The resulting string is then matched to a "white list" of allowed plates, to check the transit authorization. In case matching fails, a neuro-fuzzy agent analyses the features of the mismatch, to filter out possible false alarms and to optimize the workload for the human supervisor that has to validate the alarm.