10-04-2017, 08:57 PM
Presented by:Sorin Draghici
Dept. of Computer Science
Wayne State University
A neural network based artificial vision system for licence plate recognition
Abstract
This paper presents a neural network based artificial vision system able to analyse the image of a car given by a camera, locate the registration plate and recognise the registration number of the car. The paper describes in detail various practical problems encountered in implementing this particular application and the solutions used to solve them. The main features of the system presented are: controlled stability-plasticity behaviour, controlled reliability threshold, both off-line and on-line learning, self assessment of the output reliability and high reliability based on high level multiple feedback. The system has been designed using a modular approach which allows easy upgrading and/or substituting of various sub-modules thus making it potentially suitable in a large range of vision applications. The OCR engine was designed as an interchangeable plug-in module. This allows the user to choose an OCR engine which is suited to the particular application and to upgrade it easily in the future. At present, there are several versions of this OCR engine. One of them is based on a fully connected feedforward artificial neural network with sigmoidal activation functions. This network can be trained with various training algorithms such as error backpropagation. An alternative OCR engine is based on the constraint based decomposition (CBD) training architecture. The system has showed the following performances (on average) on real-world data: successful plate location and segmentation about 99%, successful character recognition about 98% and successful recognition of complete registration plates about 80%. Keywords: vision, OCR, neural networks, car licence plate, real-world application
Introduction
This paper presents a neural network based artificial vision system -Visicar - and its applications. The system is able to analyse the image of a car given by a camera, locate the registration plate and recognise the registration number of the car. The paper describes in details various practical problems encountered in implementing this particular application and the methods used to solve them. The paper is structured as follows. Section 2 presents some problems which create the need for such a system. Section 3 describes the system and is divided into two sub-sections. Section 3.1 presents the structure of the system whereas section 3.2 gives a detailed description of the processing performed by the system. Section 4 describes briefly the performances achieved by the system and section 5 presents some conclusions.
Security problems which create the need for such a system
This section describes briefly some situations in which non-trivial security problems can be solved by using such an artificial vision system. Parking areas with no special security requirements. It might seem that such areas do not require any security system. In reality, fraudulent practice is rather common and brings important losses to companies which manage parking areas and garages. A common fraudulent practice is to use two cars in order to occupy permanently a space in a parking lot. One can enter in the car park with a car A (a Ferrari for instance) and obtain a ticket TA stamped with the time of entrance T1. At any later date, the same person can enter with a car B (an old Mini for instance) and obtain a ticket TB stamped with the time of entrance T2. Then, the person can leave car B in the car park, and leave the car park at time T2+ with car A and ticket TB, paying just the minimum amount due for the time . Later on, car A will be deposited again in the car park with a ticket TA which will be used to exit the car park with car B (paying again just a minimum fee). The process is then repeated, always swapping cars and exiting the car park with the most recent ticket. Thus, an expensive Ferrari can be kept in a safe car park for unlimited lengths of time, almost free, with huge losses for the car park company. Another typical situation is that of a car theft. A thief can enter a car park with their own car A obtaining a ticket TA, steal a very expensive car B and leave quietly with the stolen car and the ticket TA. This type of fraud brings huge losses for car park companies materialised in high insurance costs. One can imagine a system which recognises automatically the car number plate when the car enters the parking area and stores somehow the registration number on the ticket. Later, when the car leaves the parking lot, the system can check the correspondence between the information on the ticket and the registration number of the car. It is easy to appreciate that such a system would eliminate completely both fraud situations described above or at least, reduce their number. Parking areas with security requirements. In these situations, such a system adds a further level of security by granting entrance only to registered vehicles. Toll payment. A system able to recognise registration plates can be used to identify vehicles which transit through the toll gates. Such a system can be used to achieve two types of goals. Firstly, the system can be used in conjunction with a database containing registration data and owners information in order to debit the amount due directly into the car owner s account. This can greatly reduce the running costs of the toll bridge or motorway by reducing or eliminating the need for a human presence. Secondly, such a system can be used as a back-up system which deals only with fraudulent vehicles. For instance, in Italy, the motorway system is run by a private company called Autostrade spa . This company has perfected a remote sensing system called Telepass which is able to identify certain vehicles which are fitted with a special device. Those vehicles are allowed to transit without stopping through certain dedicated channels at the toll gates, thus eliminating queuing. However, fraudulent users can transit those dedicated
For more information about this article,please follow the link:
http://vortex.cs.wayne.edu/papers/ijns1997.pdf