即時動態車牌辨識

Dynamic Real-Time License Plate Recognition

10.29507/JT.201006.0007

李建興(Chien-Hsing Lee)；游凱倫(Kai-Lun You)；林應璞(Ying-Pu Lin)

連通標記法 ； 投影量切割法 ； 雷登轉換法 ； 改良式主成分分析法 ； 歐式距離法 ； 相似符號 ； connected component analysis ； projection histogram method ； Radon transform ； improved principal component analysis ； Euclidean distance ； ambiguous characters

技術學刊

25卷2期（2010 / 06 / 01）

151 - 165

繁體中文

本文使用1394攝影機拍攝車牌影像，並與電腦連結，再由Matlab/Simulink偵測車牌存在與否。當偵測到車牌時，Matlab/Simulink 會自動儲存所偵測之車牌影像，再開啟所設計的GUI辨識介面，以讀取所儲存的車牌影像，再執行車牌定位、車牌傾斜矯正、字元分割及字元辨識等步驟。首先，以連通標記法搭配車牌區域分析來偵測車牌位置，如所拍攝車牌傾斜時，再以雷登轉換來矯正車牌，進而以投影切割法分割車牌字元。於車牌辨識率部份，以改良式主成份分析法擷取字元特徵，再搭配歐式距離決策法，以探討搭配使用及未使用簡單分類器之車牌辨識的差異。由結果得知，搭配使用與未使用簡單分類器之字元辨識成功率分別爲95.9%與87.4%左右，而其辨識時間分別爲1.3 秒／張與0.8秒／張，所測試影像皆爲640×480大小。而因字元「B」和「8」、「D」和「0」、「5」和「S」、「2」和「Z」及「1」和「I」的字型相當相似，爲改善相似符號難以辨別的問題，吾人利用字元邊緣線特徵的差異，作爲相似符號確認的依據。

This paper presents a dynamic real-time license plate recognition method using a 1394-based digital camera to record an image of a license plate and process the image with Matlab/Simulink based on GUI (graphical user interface). If the program detects a license plate in the image, the recognition system will automaticlly store the image and execute license plate localization, license plate slant correction, character segmentation and character recognition. First, the license plate in the image is localized by applying connected component analysis with the aid of license plate area analysis. If the localized license plate is slanted, the plate slant is corrected using Radon transform and its plate character is then segmented by projection histogram. As for license plate recognition, an improved principal component analysis using the Euclidean distance method with and without the aid of using a simple classifier to extract the features of the plate characters and the classification of the plate is investigated in this paper. As a result, the recognition rates of license plates for 640×480 pixel images are approximately 95.9% and 87.4% with and without the aid of a simple classifier, respectively. Their corresponding recognition times per license plate are about 1.3 and 0.8 seconds, respectively. Moreover, recognition of ambiguous characters on license plates such as the sets of ”B” and ”8”, ”D” and ”0”, ”5” and ”S”, ”2” and ”Z” as well as ”1” and ”I” are analyzed by using the difference between their edges to increase the recognition rate.

1. 洪膺倫(2017)。手搖飲料業自動化導入人工智慧系統之探討。中原大學國際經營與貿易學系學位論文。2017。1-54。