車牌字元粹取

Abstract

自動車牌辨識系統的應用相當廣泛，例如大樓車輛進出管理，電子公路收費系統，停車場收費管理等。一個辨識率高而且可靠的系統，不但可以做到無人控管，還可以節省人工管理費用。 車牌字元淬取與車牌字元辨識為本車牌自動辨識系統中的兩大核心部分，本論文以前級的車牌字元淬取為主，利用影像處理，將正確的車牌字元淬取出來，交給後級的車牌字元辨識系統。 車牌字元淬取流程分為三部份：前處理，物件搜尋和字元物件判斷。在前處理中，我們用改良式的二值化法，將車牌字元與車牌分離；接著，我們找出所有可能是車牌字元的相連物件，並標記其位置、大小及屬性；最後利用每一個物件的位置、大小和屬性值，決定正確的車牌字元物件。這樣的方式，較不會受車牌大小、方位的影響，對於複雜背景也有很好的效果。 在實際架設的場地中，統計238輛汽車測試的結果，成功淬取出車牌字元的比率為百分之93.7 ，若不考慮加框車牌，成功率甚至可達到百分之95.5 。

An automatic vehicle recognition system is expected to have numerous applications, e.g. building automation and security, electric payment toll-gate, car parking system, and etc. An automatic system with high recognition accuracy and reliability will achieve the goal of unwatched management and save the wage of guards. The extraction and recognition of vehicle plate characters are two principle stages of our automatic system. This research centers on the prior-stage, the extraction of vehicle plate characters, and extract the vehicle plate characters correctly to the post-stage, the recognition system, by using image processing technology. We divide the process of extraction in three parts: preprocessing, connected objects finding and character-objects judgement. In pre-processing, we use modified thresholding to separate the vehicle plate character from its plate. Then we find all candidate character connected objects and record its position, size and other properties. Finally, we judge the correct character objects by using the position, size and properties of all connected objects. The process we provided is highly tolerant of skewed plate and complex background. In a building parking lots, we test 238 vehicles. The percentaage of the extraction rate is 93.7 . And if we exclude the car with frame on plate, the percentage of the extraction rate can even reach 93.7 .