葉榮木蔡俊明許溥鑫2019-09-032008-9-52019-09-032007http://etds.lib.ntnu.edu.tw/cgi-bin/gs32/gsweb.cgi?o=dstdcdr&s=id=%22GN0691730139%22.&%22.id.&http://rportal.lib.ntnu.edu.tw:80/handle/20.500.12235/97133交通工具數量的激增是現代化國家均有的問題之ㄧ，在台灣這地窄人稠的地方，更突顯出此問題的嚴重性。根據國際道路聯盟所做的統計，在台灣平均每戶家庭小客車數有0.77台為亞洲之冠，其中平均每79.81輛車就有1台發生事故，完全反映了交通工具數量增加所造成的交通問題。在警察數量遠不及車輛數目的今日，如何對交通工具做更有效的管理是刻不容緩的問題。 現今之車牌定位系統，大多以白天且有夠多亮度的影像進行定位，對於亮度不均勻（包含黃昏或夜晚）之影像卻沒有進行車牌定位研究。因此，本文利用自動化影像處理，針對不同亮度下（包含日、夜間）的汽、機車影像，進行車牌定位。本文首先提出利用平均亮度值來區分白天與夜間的車牌偵測之演算法，再利用高增幅濾波器（High Boost Filter）來增強影像對比，接著，利用形態梯度（Morphology Gradient）運算子去偵測車牌區域，接著，利用平均亮度值乘上固定常數作為二值化閥值，進行影像二值化，再利用車牌外觀特性完成車牌初步定位。 本文主要針對兩車道之影像進行車牌定位，同時考量車牌解析度過低的問題，故使用640×480 pixel之影像。在白天部分，汽車車牌定位成功率為96.2％，機車車牌定位成功率為95.4％，在夜間部份，汽車車牌定位成功率為88.5，機車車牌定位成功率為75.7％。白天車牌定位總成功率為95.9%，速度為0.25sec，夜間車牌定位總成功率為83% ，速度為0.36sec。The increasing amount of vehicles is one of the common problems for most developed countries, and it is more critical for Taiwan which the population density is higher than lots of countries. According to the International Road Federation(IRF) census in 2005, in Taiwan, every family has 0.77 car in average, and there is a car got accident in every 79.81 cars. The above data totally reflect the problems made by the increasing numbers of vehicles. For nowadays the number of police far less than the vehicle, how to effectively manage the vehicle is an urgent issue. Today, most license plate location systems are applied for daytime image, which has enough brightness. However, they can’t be applied for dark and nighttime images. In this paper, we proposed a algorithm to locate license plate of cars and motorcycles under day and night conditions. At first, we calculated the value of average brightness of picture to distinguish day and night. After intensifying the picture contrast by a high boost filter, we apply the morphology gradient for detecting the plate candidate. Furthermore, we multiplied a constant to the average brightness value of images to be the threshold to image binarization, and checked the characteristic of plate appearance and calculated the three horizontal lines crossings to extract the correct plate position. The experiment results show that the average extraction rate and speed are 95.9%, 0.25sec at daytime and 83%, 0.36sec at nighttime. We believe the proposed method would be a robust license plate extraction system utilized in all day.影像增強形態梯度二值化車牌定位Image EnhancementMorphology GradientBinarizationLocation of License Plate