關於車牌辨識系統的研究已經有一段時間了,但對於機車的辨識、多目標辨識與車輛顏色識別的研究卻不多見。有些國家使用機車的人數比汽車還多,因此機車的辨識也是相當重要。由於拍攝的照片裡,可能含有一部以上的車輛,因此必須一一辨識。拍攝環境如果是在室外,則環境背景會較為複雜,如樹木、招牌、交通號誌等物體都會造成車牌定位上的困難,因此必須克服背景複雜的問題。 本研究提出了具有適應環境背景複雜能力的汽機車車牌的辨識、多目標的辨識與汽車車輛顏色的識別。車牌辨識系統主要分為車牌定位、字元切割與字元辨識等三大部分。本研究使用了894張樣本進行實驗,結果在單目標的車牌定位率平均達97.65%,多目標的車牌定位率平均達96.59%,字元辨識率平均達98.02%,而車輛顏色識別率平均則達82.96%。
The study of vehicle license plate recognition has been undergone for decades, but studies for identifying vehicle colors, license plates for multi-target vehicles, and license plates for motorcycles are still being greatly missed. It is considered important to study the license plate recognition also for motorcycles since in many countries the number of motorcycles is even more than that of cars. Since the pictures taken by surveillance cameras hardly contain only a single vehicle, it is inevitable to analyze and identify all the vehicles appeared in the picture. Moreover, if the pictures take outside, it is also known for their complex background objects such as trees, billboards, traffic signs, etc. which will impose difficulties for license plate recognition. In this study, a system is proposed for multi-target vehicle license plate recognition. This system is capable of recognizing license plate as well as the color of a vehicle with a complex background. The vehicle license plate recognition system may be divided into three major components, namely, license plate detection, character segmentation, and character recognition. To demonstrate the performance of the system, there are 894 different images used for testing. The average accuracy is 97.65% for the single-target license plate detection and 96.59% for the multi-target license plate detection. In addition, the average accuracy is 98.02% for the character recognition phase, and 82.96% for car colors classification phase.