License plate survey for traffic analysis: improving accuracy with correction algorithms

Abstract

Vehicle tracking methods are widely used for a variety of purposes including collection of travel time and duration of stay data. The collected data are used for planning and management purposes. The type of data depends on the method of data collection. Tracking methods are usually classified into active and passive. In this research they are classified into two categories, discrete and continuous. Among all methods, the discrete method of license plate matching is the most prevalent for data collection. The purpose of this research is to discuss the accuracy of manual license plate matching method for vehicle tracking and travel time data collection, and provide correction algorithms to improve the results. The impacts of recordation style and visual similarities between characters (letters and numbers) on the matching errors are investigated. The correction algorithms are compared and evaluated. The application of correction algorithms--specifically those that are more constrained to filter out false matches--can considerably increase the percentage of matched license plates. To a lesser degree, this processing can improve the statistical values of the license plate datasets such as average, standard deviation and median of travel time and duration of stay in a location. This study also found evidence that a significant portion of mistakenly recorded letters while recording the license plates are visually similar letters, that by itself underlines the human factor in the accuracy of the method. Digits are not significantly probable to be mistaken because of their visual dissimilarity. The workload of recordation is also proved to be significant: more letters to be recorded results in more errors.