This paper describes how a number plate recogniser developed by Computer Recognition Systems has been evaluated for transport planning purposes. Both laboratory and field trials have been undertaken. Topics covered include a detailed statistical analysis of the error rates associated with the different number plate syntaxes and character sets currently in use, along with detailed measures of the accuracy with which each alphanumeric character is deciphered under a wide variety of conditions. From this information statistical procedures for matching complete and partial registration data and for coping with spurious matches in origin destination, journey time and entry- exit flows in a network are being devised that are appropriate to the recogniser characteristics. This is being achieved by taking into account the error characteristics of the survey technique, which in this case involves a probabilistic analysis of the possible errors. The physical issues involved in using automated registration plate recognition are also being studied such as the choice of camera equipment, location of test sites and weather conditions.
Abstract